diff --git a/test/registered/4-gpu-models/test_gpt_oss_4gpu.py b/test/registered/4-gpu-models/test_gpt_oss_4gpu.py
index c31deb720..58c71e72b 100644
--- a/test/registered/4-gpu-models/test_gpt_oss_4gpu.py
+++ b/test/registered/4-gpu-models/test_gpt_oss_4gpu.py
@@ -3,8 +3,8 @@ import unittest
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.gpt_oss_common import BaseTestGptOss
 
-register_cuda_ci(est_time=328, suite="stage-c-test-4-gpu-h100")
-register_cuda_ci(est_time=740, suite="stage-c-test-4-gpu-b200-small")
+register_cuda_ci(est_time=392, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=584, suite="stage-c-test-4-gpu-b200-small")
 
 
 class TestGptOss4Gpu(BaseTestGptOss):
diff --git a/test/registered/4-gpu-models/test_qwen35_fp4_mtp_v2.py b/test/registered/4-gpu-models/test_qwen35_fp4_mtp_v2.py
index 53c8e7917..1f7034286 100644
--- a/test/registered/4-gpu-models/test_qwen35_fp4_mtp_v2.py
+++ b/test/registered/4-gpu-models/test_qwen35_fp4_mtp_v2.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=540, suite="stage-c-test-4-gpu-b200-small")
+register_cuda_ci(est_time=422, suite="stage-c-test-4-gpu-b200-small")
 
 QWEN35_FP4_MODEL = "nvidia/Qwen3.5-397B-A17B-NVFP4"
 ACC_THRESHOLDS = {QWEN35_FP4_MODEL: {"gsm8k": 0.95}}
diff --git a/test/registered/4-gpu-models/test_qwen35_fp4_triton.py b/test/registered/4-gpu-models/test_qwen35_fp4_triton.py
index 29b7e5ce0..96363bf69 100644
--- a/test/registered/4-gpu-models/test_qwen35_fp4_triton.py
+++ b/test/registered/4-gpu-models/test_qwen35_fp4_triton.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     ModelLaunchSettings,
 )
 
-register_cuda_ci(est_time=720, suite="stage-c-test-4-gpu-b200-small")
+register_cuda_ci(est_time=563, suite="stage-c-test-4-gpu-b200-small")
 
 QWEN35_FP4_MODEL = "nvidia/Qwen3.5-397B-A17B-NVFP4"
 ACC_THRESHOLDS = {QWEN35_FP4_MODEL: {"gsm8k": 0.95}}
diff --git a/test/registered/4-gpu-models/test_qwen3_30b.py b/test/registered/4-gpu-models/test_qwen3_30b.py
index 487445a24..d2461da92 100644
--- a/test/registered/4-gpu-models/test_qwen3_30b.py
+++ b/test/registered/4-gpu-models/test_qwen3_30b.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=146, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=261, suite="stage-c-test-4-gpu-h100")
 
 QWEN3_30B_MODEL_PATH = "Qwen/Qwen3-30B-A3B-FP8"
 
diff --git a/test/registered/4-gpu-models/test_qwen3_next_models.py b/test/registered/4-gpu-models/test_qwen3_next_models.py
index c9bc1839e..d6a801d52 100644
--- a/test/registered/4-gpu-models/test_qwen3_next_models.py
+++ b/test/registered/4-gpu-models/test_qwen3_next_models.py
@@ -6,7 +6,7 @@ from sglang.test.kits.kl_divergence_kit import KLDivergenceMixin
 from sglang.test.kits.prefix_cache_branching_kit import PrefixCacheBranchingMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=119, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=142, suite="stage-c-test-4-gpu-h100")
 
 QWEN3_NEXT_MODEL = "Qwen/Qwen3-Next-80B-A3B-Instruct"
 
diff --git a/test/registered/8-gpu-models/test_deepseek_v32_indexcache.py b/test/registered/8-gpu-models/test_deepseek_v32_indexcache.py
index 320e23028..e9b3d8633 100644
--- a/test/registered/8-gpu-models/test_deepseek_v32_indexcache.py
+++ b/test/registered/8-gpu-models/test_deepseek_v32_indexcache.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=354, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=492, suite="stage-c-test-8-gpu-h200")
 
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 
diff --git a/test/registered/8-gpu-models/test_deepseek_v3_basic.py b/test/registered/8-gpu-models/test_deepseek_v3_basic.py
index c814ed127..cd0bc5c2c 100644
--- a/test/registered/8-gpu-models/test_deepseek_v3_basic.py
+++ b/test/registered/8-gpu-models/test_deepseek_v3_basic.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=320, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=301, suite="stage-c-test-8-gpu-h200")
 
 FULL_DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
 
diff --git a/test/registered/8-gpu-models/test_deepseek_v3_mtp.py b/test/registered/8-gpu-models/test_deepseek_v3_mtp.py
index 805db513d..8089d1167 100644
--- a/test/registered/8-gpu-models/test_deepseek_v3_mtp.py
+++ b/test/registered/8-gpu-models/test_deepseek_v3_mtp.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=198, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=309, suite="stage-c-test-8-gpu-h200")
 
 FULL_DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
 
diff --git a/test/registered/8-gpu-models/test_dsa_models_basic.py b/test/registered/8-gpu-models/test_dsa_models_basic.py
index cc6caffd2..ca1ebcd75 100644
--- a/test/registered/8-gpu-models/test_dsa_models_basic.py
+++ b/test/registered/8-gpu-models/test_dsa_models_basic.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=966, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=1047, suite="stage-c-test-8-gpu-h200")
 
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 GLM5_MODEL_PATH = "zai-org/GLM-5-FP8"
diff --git a/test/registered/8-gpu-models/test_dsa_models_mtp.py b/test/registered/8-gpu-models/test_dsa_models_mtp.py
index 9c0b2dcef..fe5ffe1c7 100644
--- a/test/registered/8-gpu-models/test_dsa_models_mtp.py
+++ b/test/registered/8-gpu-models/test_dsa_models_mtp.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=910, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=1048, suite="stage-c-test-8-gpu-h200")
 
 FULL_DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 GLM5_MODEL_PATH = "zai-org/GLM-5-FP8"
diff --git a/test/registered/8-gpu-models/test_mimo_models.py b/test/registered/8-gpu-models/test_mimo_models.py
index fe79139bd..e2651ad30 100644
--- a/test/registered/8-gpu-models/test_mimo_models.py
+++ b/test/registered/8-gpu-models/test_mimo_models.py
@@ -5,7 +5,7 @@ from sglang.test.kits.eval_accuracy_kit import GSM8KMixin
 from sglang.test.kits.spec_decoding_kit import SpecDecodingMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=283, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=330, suite="stage-c-test-8-gpu-h200")
 
 
 class TestMiMoV2Flash(GSM8KMixin, SpecDecodingMixin, DefaultServerBase):
diff --git a/test/registered/8-gpu-models/test_minimax_m25_basic.py b/test/registered/8-gpu-models/test_minimax_m25_basic.py
index f99e4c3cd..894dfe89b 100644
--- a/test/registered/8-gpu-models/test_minimax_m25_basic.py
+++ b/test/registered/8-gpu-models/test_minimax_m25_basic.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=330, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=307, suite="stage-c-test-8-gpu-h200")
 
 MINIMAX_M25_MODEL_PATH = "MiniMaxAI/MiniMax-M2.5"
 
diff --git a/test/registered/8-gpu-models/test_nvidia_nemotron_3_super_bf16.py b/test/registered/8-gpu-models/test_nvidia_nemotron_3_super_bf16.py
index 5d2200f3c..0e35999b3 100644
--- a/test/registered/8-gpu-models/test_nvidia_nemotron_3_super_bf16.py
+++ b/test/registered/8-gpu-models/test_nvidia_nemotron_3_super_bf16.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=319, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=376, suite="stage-c-test-8-gpu-h200")
 
 NEMOTRON_3_SUPER_BF16_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"
 
diff --git a/test/registered/8-gpu-models/test_step3p5_flash_chain_mtp.py b/test/registered/8-gpu-models/test_step3p5_flash_chain_mtp.py
index 40d188180..15abe79ce 100644
--- a/test/registered/8-gpu-models/test_step3p5_flash_chain_mtp.py
+++ b/test/registered/8-gpu-models/test_step3p5_flash_chain_mtp.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=500, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=663, suite="stage-c-test-8-gpu-h200")
 
 STEP3P5_FLASH_MODEL_PATH = "stepfun-ai/Step-3.5-Flash"
 
diff --git a/test/registered/attention/test_chunk_gated_delta_rule.py b/test/registered/attention/test_chunk_gated_delta_rule.py
index f1ac1a40f..e9f9c7ad5 100644
--- a/test/registered/attention/test_chunk_gated_delta_rule.py
+++ b/test/registered/attention/test_chunk_gated_delta_rule.py
@@ -8,7 +8,7 @@ from sglang.srt.layers.attention.fla.fused_recurrent import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=5, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "Test requires CUDA")
diff --git a/test/registered/attention/test_fa3.py b/test/registered/attention/test_fa3.py
index b941e134d..87fd59ac4 100644
--- a/test/registered/attention/test_fa3.py
+++ b/test/registered/attention/test_fa3.py
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
 
 # FlashAttention3 integration tests (requires SM 90+ / H100)
 # Multiple test classes: FA3, FA3+MLA, FA3+SpecDecode variants
-register_cuda_ci(est_time=386, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=551, suite="stage-b-test-1-gpu-large")
 
 GSM_DATASET_PATH = None
 
diff --git a/test/registered/attention/test_flash_attention_4.py b/test/registered/attention/test_flash_attention_4.py
index 83a34c968..229096307 100644
--- a/test/registered/attention/test_flash_attention_4.py
+++ b/test/registered/attention/test_flash_attention_4.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
 )
 
 # FlashAttention4 integration test (requires SM 100+ / Blackwell B200)
-register_cuda_ci(est_time=332, suite="stage-b-test-4-gpu-b200")
+register_cuda_ci(est_time=265, suite="stage-b-test-4-gpu-b200")
 
 
 @unittest.skipIf(get_device_sm() < 100, "Test requires CUDA SM 100 or higher")
diff --git a/test/registered/attention/test_gdn_noncontiguous_stride.py b/test/registered/attention/test_gdn_noncontiguous_stride.py
index 81b28f66d..af807978b 100644
--- a/test/registered/attention/test_gdn_noncontiguous_stride.py
+++ b/test/registered/attention/test_gdn_noncontiguous_stride.py
@@ -14,7 +14,7 @@ from sglang.srt.layers.attention.fla.fused_sigmoid_gating_recurrent import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=30, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=7, suite="stage-b-test-1-gpu-large")
 
 
 def _make_noncontiguous_ab(batch, num_heads, dtype=torch.bfloat16, device="cuda"):
diff --git a/test/registered/attention/test_hybrid_attn_backend.py b/test/registered/attention/test_hybrid_attn_backend.py
index dce9ca3c6..d99943842 100644
--- a/test/registered/attention/test_hybrid_attn_backend.py
+++ b/test/registered/attention/test_hybrid_attn_backend.py
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
 
 # Hybrid attention backend tests (FA3 prefill + FlashInfer decode, requires SM 90+ / H100)
 # Multiple test classes: base, MLA, TorchCompile, SpecDecode variants
-register_cuda_ci(est_time=342, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=407, suite="stage-b-test-1-gpu-large")
 
 GSM_DATASET_PATH = None
 
diff --git a/test/registered/attention/test_kda_kernels.py b/test/registered/attention/test_kda_kernels.py
index 51f4e9db1..42bdea4f0 100644
--- a/test/registered/attention/test_kda_kernels.py
+++ b/test/registered/attention/test_kda_kernels.py
@@ -13,7 +13,7 @@ from sglang.srt.layers.attention.fla.kda import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=12, suite="stage-b-test-1-gpu-large")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "Test requires CUDA")
diff --git a/test/registered/attention/test_local_attn.py b/test/registered/attention/test_local_attn.py
index f14be8cf7..f3c0a2c6e 100644
--- a/test/registered/attention/test_local_attn.py
+++ b/test/registered/attention/test_local_attn.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
 )
 
 # Local attention with FA3 (requires SM 90+ / H100, tp=4)
-register_cuda_ci(est_time=208, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=217, suite="stage-c-test-4-gpu-h100")
 
 
 @unittest.skipIf(get_device_sm() < 90, "Test requires CUDA SM 90 or higher")
diff --git a/test/registered/attention/test_normal_decode_set_metadata.py b/test/registered/attention/test_normal_decode_set_metadata.py
index 6906f6cfa..b4c6e5c12 100644
--- a/test/registered/attention/test_normal_decode_set_metadata.py
+++ b/test/registered/attention/test_normal_decode_set_metadata.py
@@ -21,7 +21,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 # Register this test for CUDA CI in stage-b (fast attention/kernel tests)
-register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-large")
 
 
 def reference_normal_decode_set_metadata(
diff --git a/test/registered/attention/test_torch_native_attention_backend.py b/test/registered/attention/test_torch_native_attention_backend.py
index 102a7816b..eba008c64 100644
--- a/test/registered/attention/test_torch_native_attention_backend.py
+++ b/test/registered/attention/test_torch_native_attention_backend.py
@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
 )
 
 # Torch native attention backend integration test with MMLU eval
-register_cuda_ci(est_time=128, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=140, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=150, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/attention/test_triton_attention_backend.py b/test/registered/attention/test_triton_attention_backend.py
index 2034a89a6..e8558d13c 100644
--- a/test/registered/attention/test_triton_attention_backend.py
+++ b/test/registered/attention/test_triton_attention_backend.py
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
 )
 
 # Triton attention backend integration test with latency benchmark and MMLU eval
-register_cuda_ci(est_time=164, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=177, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=1400, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/attention/test_triton_attention_kernels.py b/test/registered/attention/test_triton_attention_kernels.py
index 48b74cb72..9dde8c16d 100644
--- a/test/registered/attention/test_triton_attention_kernels.py
+++ b/test/registered/attention/test_triton_attention_kernels.py
@@ -23,7 +23,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase, is_in_amd_ci
 
 # Triton attention kernel unit tests (decode, extend, prefill)
-register_cuda_ci(est_time=16, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=19, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=30, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/attention/test_triton_sliding_window.py b/test/registered/attention/test_triton_sliding_window.py
index faff86108..42aa770e3 100644
--- a/test/registered/attention/test_triton_sliding_window.py
+++ b/test/registered/attention/test_triton_sliding_window.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
 )
 
 # Sliding window attention with Triton backend (Gemma-3 model)
-register_cuda_ci(est_time=95, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=93, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=200, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/backends/test_torch_compile.py b/test/registered/backends/test_torch_compile.py
index 3f5f1ca91..d884631f9 100644
--- a/test/registered/backends/test_torch_compile.py
+++ b/test/registered/backends/test_torch_compile.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=115, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=126, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=1100, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/breakable_cuda_graph/test_breakable_cuda_graph.py b/test/registered/breakable_cuda_graph/test_breakable_cuda_graph.py
index 588f3e91f..e69674dd7 100644
--- a/test/registered/breakable_cuda_graph/test_breakable_cuda_graph.py
+++ b/test/registered/breakable_cuda_graph/test_breakable_cuda_graph.py
@@ -23,7 +23,7 @@ from sglang.test.test_utils import (
 )
 
 # CI Registration — large suite to fit the integration test's server startup.
-register_cuda_ci(est_time=130, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=79, suite="stage-b-test-1-gpu-large")
 
 
 def _skip_if_no_cuda(test_func):
diff --git a/test/registered/constrained_decoding/test_constrained_decoding.py b/test/registered/constrained_decoding/test_constrained_decoding.py
index 9e3cf9213..9a71f75fd 100644
--- a/test/registered/constrained_decoding/test_constrained_decoding.py
+++ b/test/registered/constrained_decoding/test_constrained_decoding.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=110, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=179, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/core/test_deterministic.py b/test/registered/core/test_deterministic.py
index 7b269a93f..f28db8d0b 100644
--- a/test/registered/core/test_deterministic.py
+++ b/test/registered/core/test_deterministic.py
@@ -16,7 +16,7 @@ from sglang.test.test_deterministic_utils import (
 )
 from sglang.test.test_utils import is_in_amd_ci
 
-register_cuda_ci(est_time=194, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=207, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=278, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/core/test_engine_child_pids.py b/test/registered/core/test_engine_child_pids.py
index 7762ef79e..eaa103e40 100644
--- a/test/registered/core/test_engine_child_pids.py
+++ b/test/registered/core/test_engine_child_pids.py
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
-register_cuda_ci(est_time=60, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=77, suite="stage-b-test-1-gpu-small")
 
 
 class TestEngineChildPids(CustomTestCase):
diff --git a/test/registered/core/test_gpt_oss_1gpu.py b/test/registered/core/test_gpt_oss_1gpu.py
index 692af9070..c6b6e5162 100644
--- a/test/registered/core/test_gpt_oss_1gpu.py
+++ b/test/registered/core/test_gpt_oss_1gpu.py
@@ -3,7 +3,7 @@ import unittest
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.gpt_oss_common import BaseTestGptOss
 
-register_cuda_ci(est_time=372, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=408, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=750, suite="stage-b-test-1-gpu-small-amd-mi35x")
 
 
diff --git a/test/registered/core/test_gpt_oss_sm120.py b/test/registered/core/test_gpt_oss_sm120.py
index 18ad85478..d453d3dab 100644
--- a/test/registered/core/test_gpt_oss_sm120.py
+++ b/test/registered/core/test_gpt_oss_sm120.py
@@ -5,7 +5,7 @@ import torch
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.gpt_oss_common import BaseTestGptOss
 
-register_cuda_ci(est_time=329, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=345, suite="stage-b-test-1-gpu-small")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is not available")
diff --git a/test/registered/core/test_hidden_states.py b/test/registered/core/test_hidden_states.py
index 1e8d8a764..2f83fea90 100644
--- a/test/registered/core/test_hidden_states.py
+++ b/test/registered/core/test_hidden_states.py
@@ -8,7 +8,7 @@ from sglang.srt.utils import get_device, is_hip
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST, CustomTestCase
 
-register_cuda_ci(est_time=47, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=45, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=55, suite="stage-b-test-1-gpu-small-amd")
 
 _is_hip = is_hip()
diff --git a/test/registered/core/test_mm_process_config.py b/test/registered/core/test_mm_process_config.py
index 35540d8d7..aef4fd3c5 100644
--- a/test/registered/core/test_mm_process_config.py
+++ b/test/registered/core/test_mm_process_config.py
@@ -4,7 +4,7 @@ from unittest.mock import MagicMock, patch
 from sglang.srt.server_args import ServerArgs
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=5, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=1, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/core/test_request_queue_validation.py b/test/registered/core/test_request_queue_validation.py
index d6293ae2f..82147c41f 100644
--- a/test/registered/core/test_request_queue_validation.py
+++ b/test/registered/core/test_request_queue_validation.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     send_generate_requests,
 )
 
-register_cuda_ci(est_time=50, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=53, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=70, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/core/test_srt_endpoint.py b/test/registered/core/test_srt_endpoint.py
index a639d7861..081c93cb7 100644
--- a/test/registered/core/test_srt_endpoint.py
+++ b/test/registered/core/test_srt_endpoint.py
@@ -27,7 +27,7 @@ from sglang.test.test_utils import (
     run_logprob_check,
 )
 
-register_cuda_ci(est_time=132, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=134, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=130, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/core/test_srt_engine.py b/test/registered/core/test_srt_engine.py
index f6610ddf8..2d769631a 100644
--- a/test/registered/core/test_srt_engine.py
+++ b/test/registered/core/test_srt_engine.py
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
-register_cuda_ci(est_time=311, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=387, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=261, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/cp/test_deepseek_v32_cp_single_node.py b/test/registered/cp/test_deepseek_v32_cp_single_node.py
index badcb2caa..5adefd3d1 100644
--- a/test/registered/cp/test_deepseek_v32_cp_single_node.py
+++ b/test/registered/cp/test_deepseek_v32_cp_single_node.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=614, suite="stage-c-test-deepep-8-gpu-h200")
+register_cuda_ci(est_time=616, suite="stage-c-test-deepep-8-gpu-h200")
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 
 
diff --git a/test/registered/disaggregation/test_disaggregation_basic.py b/test/registered/disaggregation/test_disaggregation_basic.py
index e20d59ea8..fdd62ce90 100644
--- a/test/registered/disaggregation/test_disaggregation_basic.py
+++ b/test/registered/disaggregation/test_disaggregation_basic.py
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     DEFAULT_TARGET_MODEL_EAGLE3,
 )
 
-register_cuda_ci(est_time=394, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=509, suite="stage-b-test-2-gpu-large")
 
 
 class TestDisaggregationAccuracy(PauseResumeInPlaceMixin, PDDisaggregationServerBase):
diff --git a/test/registered/distributed/test_data_parallelism.py b/test/registered/distributed/test_data_parallelism.py
index a7d49ac19..fa7ac2170 100644
--- a/test/registered/distributed/test_data_parallelism.py
+++ b/test/registered/distributed/test_data_parallelism.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=78, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=91, suite="stage-b-test-2-gpu-large")
 register_amd_ci(est_time=73, suite="stage-b-test-2-gpu-large-amd")
 
 
diff --git a/test/registered/distributed/test_disaggregation_different_tp.py b/test/registered/distributed/test_disaggregation_different_tp.py
index 3fd1a9504..f3ed6ad35 100644
--- a/test/registered/distributed/test_disaggregation_different_tp.py
+++ b/test/registered/distributed/test_disaggregation_different_tp.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=331, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=375, suite="stage-c-test-8-gpu-h20")
 
 
 class TestDisaggregationMooncakePrefillLargerTP(PDDisaggregationServerBase):
diff --git a/test/registered/distributed/test_disaggregation_dp_attention.py b/test/registered/distributed/test_disaggregation_dp_attention.py
index 3eab31142..e6348f83b 100644
--- a/test/registered/distributed/test_disaggregation_dp_attention.py
+++ b/test/registered/distributed/test_disaggregation_dp_attention.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=408, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=443, suite="stage-c-test-8-gpu-h20")
 
 
 class TestDisaggregationDPAttention(PDDisaggregationServerBase):
diff --git a/test/registered/distributed/test_disaggregation_hybrid_attention.py b/test/registered/distributed/test_disaggregation_hybrid_attention.py
index 87ef16974..439201d0e 100644
--- a/test/registered/distributed/test_disaggregation_hybrid_attention.py
+++ b/test/registered/distributed/test_disaggregation_hybrid_attention.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_pd_server,
 )
 
-register_cuda_ci(est_time=317, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=695, suite="stage-c-test-8-gpu-h200")
 
 
 @unittest.skipIf(is_in_ci(), "Temporarily disable the flaky test.")
diff --git a/test/registered/distributed/test_disaggregation_pp.py b/test/registered/distributed/test_disaggregation_pp.py
index 683da93a6..072b034ad 100644
--- a/test/registered/distributed/test_disaggregation_pp.py
+++ b/test/registered/distributed/test_disaggregation_pp.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=180, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=216, suite="stage-c-test-8-gpu-h20")
 
 
 class TestDisaggregationPrefillPPAccuracy(PDDisaggregationServerBase):
diff --git a/test/registered/distributed/test_dp_attention.py b/test/registered/distributed/test_dp_attention.py
index 4b9ab0ef5..c93ec10ce 100644
--- a/test/registered/distributed/test_dp_attention.py
+++ b/test/registered/distributed/test_dp_attention.py
@@ -25,7 +25,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=446, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=524, suite="stage-b-test-2-gpu-large")
 
 
 class TestDPAttentionDP2TP2(
diff --git a/test/registered/distributed/test_dp_attention_large.py b/test/registered/distributed/test_dp_attention_large.py
index 57e3f24ce..0a21c5ac7 100644
--- a/test/registered/distributed/test_dp_attention_large.py
+++ b/test/registered/distributed/test_dp_attention_large.py
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=230, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=245, suite="stage-c-test-4-gpu-h100")
 register_amd_ci(est_time=350, suite="stage-c-test-4-gpu-amd")
 
 
diff --git a/test/registered/distributed/test_load_weights_from_remote_instance.py b/test/registered/distributed/test_load_weights_from_remote_instance.py
index 645ecd82c..e48377f97 100644
--- a/test/registered/distributed/test_load_weights_from_remote_instance.py
+++ b/test/registered/distributed/test_load_weights_from_remote_instance.py
@@ -38,7 +38,7 @@ from sglang.utils import terminate_process
 
 mp.set_start_method("spawn", force=True)
 
-register_cuda_ci(est_time=110, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=145, suite="stage-b-test-2-gpu-large")
 register_amd_ci(est_time=72, suite="stage-b-test-2-gpu-large-amd")
 
 
diff --git a/test/registered/distributed/test_pp_single_node.py b/test/registered/distributed/test_pp_single_node.py
index 0549d00d2..13ed42509 100644
--- a/test/registered/distributed/test_pp_single_node.py
+++ b/test/registered/distributed/test_pp_single_node.py
@@ -32,7 +32,7 @@ from sglang.test.test_utils import (
     run_bench_one_batch_server,
 )
 
-register_cuda_ci(est_time=462, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=554, suite="stage-c-test-4-gpu-h100")
 register_amd_ci(est_time=650, suite="stage-c-test-4-gpu-amd")
 
 
diff --git a/test/registered/dllm/test_llada2_mini.py b/test/registered/dllm/test_llada2_mini.py
index 139fc2b7a..e56d3dcbd 100644
--- a/test/registered/dllm/test_llada2_mini.py
+++ b/test/registered/dllm/test_llada2_mini.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=116, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=139, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=330, suite="stage-b-test-1-gpu-small-amd")
 
 import unittest
diff --git a/test/registered/ep/test_deepep_large.py b/test/registered/ep/test_deepep_large.py
index aa9d37ec2..5967d489e 100644
--- a/test/registered/ep/test_deepep_large.py
+++ b/test/registered/ep/test_deepep_large.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=462, suite="stage-c-test-deepep-8-gpu-h200")
+register_cuda_ci(est_time=528, suite="stage-c-test-deepep-8-gpu-h200")
 
 DEEPSEEK_V32_MODEL_PATH = "deepseek-ai/DeepSeek-V3.2"
 
diff --git a/test/registered/ep/test_deepep_small.py b/test/registered/ep/test_deepep_small.py
index 1f4e8a619..0ed244774 100644
--- a/test/registered/ep/test_deepep_small.py
+++ b/test/registered/ep/test_deepep_small.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=432, suite="stage-c-test-deepep-4-gpu-h100")
+register_cuda_ci(est_time=478, suite="stage-c-test-deepep-4-gpu-h100")
 
 
 class TestPureDP(CustomTestCase):
diff --git a/test/registered/ep/test_mooncake_ep_small.py b/test/registered/ep/test_mooncake_ep_small.py
index 80f328c0d..61e0cd04e 100644
--- a/test/registered/ep/test_mooncake_ep_small.py
+++ b/test/registered/ep/test_mooncake_ep_small.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=77, suite="stage-c-test-deepep-4-gpu-h100")
+register_cuda_ci(est_time=82, suite="stage-c-test-deepep-4-gpu-h100")
 
 ib_devices = get_rdma_devices_args()
 
diff --git a/test/registered/eval/test_eval_accuracy_large.py b/test/registered/eval/test_eval_accuracy_large.py
index 7ac092d8a..c17127d48 100644
--- a/test/registered/eval/test_eval_accuracy_large.py
+++ b/test/registered/eval/test_eval_accuracy_large.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=534, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=496, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=420, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/input_embedding/test_input_embeddings.py b/test/registered/input_embedding/test_input_embeddings.py
index 373be99dd..1616fe942 100644
--- a/test/registered/input_embedding/test_input_embeddings.py
+++ b/test/registered/input_embedding/test_input_embeddings.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=41, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=42, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=38, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/input_embedding/test_input_embeds_chunked.py b/test/registered/input_embedding/test_input_embeds_chunked.py
index 963a205ce..61c3ca501 100644
--- a/test/registered/input_embedding/test_input_embeds_chunked.py
+++ b/test/registered/input_embedding/test_input_embeds_chunked.py
@@ -30,7 +30,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=42, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=43, suite="stage-b-test-1-gpu-small")
 
 CHUNKED_PREFILL_SIZE = 256
 
diff --git a/test/registered/kernels/test_nsa_indexer.py b/test/registered/kernels/test_nsa_indexer.py
index 77c007dad..9aaba4568 100644
--- a/test/registered/kernels/test_nsa_indexer.py
+++ b/test/registered/kernels/test_nsa_indexer.py
@@ -24,7 +24,7 @@ from sglang.srt.model_executor.forward_batch_info import ForwardBatch, ForwardMo
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=15, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=18, suite="stage-b-test-1-gpu-large")
 
 # Global configuration for all indexer tests
 DEFAULT_CONFIG = {
diff --git a/test/registered/language/test_srt_backend.py b/test/registered/language/test_srt_backend.py
index 551ec8103..23567f119 100644
--- a/test/registered/language/test_srt_backend.py
+++ b/test/registered/language/test_srt_backend.py
@@ -20,7 +20,7 @@ from sglang.test.test_programs import (
 )
 from sglang.test.test_utils import DEFAULT_MODEL_NAME_FOR_TEST, CustomTestCase
 
-register_cuda_ci(est_time=72, suite="stage-a-test-1-gpu-small")
+register_cuda_ci(est_time=79, suite="stage-a-test-1-gpu-small")
 register_amd_ci(est_time=120, suite="stage-a-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/layers/mamba/test_causal_conv1d.py b/test/registered/layers/mamba/test_causal_conv1d.py
index bd10fe21a..24acdd774 100644
--- a/test/registered/layers/mamba/test_causal_conv1d.py
+++ b/test/registered/layers/mamba/test_causal_conv1d.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=13, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=25, suite="stage-b-test-1-gpu-small-amd")
 
 # Adapted from https://github.com/vllm-project/vllm/blob/main/tests/kernels/mamba/test_causal_conv1d.py
diff --git a/test/registered/layers/mamba/test_mamba2_mixer.py b/test/registered/layers/mamba/test_mamba2_mixer.py
index 85aadbcd7..9ff5060ba 100644
--- a/test/registered/layers/mamba/test_mamba2_mixer.py
+++ b/test/registered/layers/mamba/test_mamba2_mixer.py
@@ -15,7 +15,7 @@ from sglang.srt.distributed.parallel_state import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=28, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=32, suite="stage-b-test-2-gpu-large")
 
 NUM_GPUS = 2
 
diff --git a/test/registered/layers/mamba/test_mamba_ssm.py b/test/registered/layers/mamba/test_mamba_ssm.py
index 8c139fbdf..6bd7d5734 100644
--- a/test/registered/layers/mamba/test_mamba_ssm.py
+++ b/test/registered/layers/mamba/test_mamba_ssm.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=10, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-small-amd")
 
 # Adapted from https://github.com/vllm-project/vllm/blob/633f943e30a4444d890d26b81850f7217736f840/tests/kernels/mamba/test_mamba_ssm_ssd.py
diff --git a/test/registered/layers/mamba/test_mamba_ssm_ssd.py b/test/registered/layers/mamba/test_mamba_ssm_ssd.py
index 2e42caa9b..3b9670d7e 100644
--- a/test/registered/layers/mamba/test_mamba_ssm_ssd.py
+++ b/test/registered/layers/mamba/test_mamba_ssm_ssd.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=10, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=34, suite="stage-b-test-1-gpu-small-amd")
 
 # Adapted from https://github.com/vllm-project/vllm/blob/633f943e30a4444d890d26b81850f7217736f840/tests/kernels/mamba/test_mamba_ssm_ssd.py
diff --git a/test/registered/lora/test_fused_moe_lora_kernel.py b/test/registered/lora/test_fused_moe_lora_kernel.py
index 50587579d..91fd2f55e 100644
--- a/test/registered/lora/test_fused_moe_lora_kernel.py
+++ b/test/registered/lora/test_fused_moe_lora_kernel.py
@@ -15,7 +15,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 
 # ==============================================================================
 
-register_cuda_ci(est_time=14, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=28, suite="stage-b-test-1-gpu-large")
 
 
 def round_up(x, base):
diff --git a/test/registered/lora/test_lora_backend.py b/test/registered/lora/test_lora_backend.py
index 9ac9f4167..4a622bbc5 100644
--- a/test/registered/lora/test_lora_backend.py
+++ b/test/registered/lora/test_lora_backend.py
@@ -29,7 +29,7 @@ from sglang.test.lora_utils import (
 )
 from sglang.test.test_utils import CustomTestCase, is_in_ci
 
-register_cuda_ci(est_time=206, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=224, suite="stage-b-test-1-gpu-small")
 register_amd_ci(
     est_time=200,
     suite="stage-b-test-1-gpu-small-amd",
diff --git a/test/registered/lora/test_lora_eviction.py b/test/registered/lora/test_lora_eviction.py
index c7c2a20d3..940db7af0 100644
--- a/test/registered/lora/test_lora_eviction.py
+++ b/test/registered/lora/test_lora_eviction.py
@@ -23,7 +23,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.runners import SRTRunner
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=238, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=263, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=224, suite="stage-b-test-1-gpu-small-amd")
 
 PROMPTS = [
diff --git a/test/registered/lora/test_lora_moe_runner.py b/test/registered/lora/test_lora_moe_runner.py
index 437f6b1c0..253fdf1e9 100644
--- a/test/registered/lora/test_lora_moe_runner.py
+++ b/test/registered/lora/test_lora_moe_runner.py
@@ -30,7 +30,7 @@ from sglang.srt.lora.lora_moe_runners import LoRAInfo
 from sglang.srt.utils import set_random_seed
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=80, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=17, suite="stage-b-test-1-gpu-large")
 
 
 def generate_request_data(
diff --git a/test/registered/lora/test_lora_overlap_loading.py b/test/registered/lora/test_lora_overlap_loading.py
index 3812b8c25..a733d4a7c 100644
--- a/test/registered/lora/test_lora_overlap_loading.py
+++ b/test/registered/lora/test_lora_overlap_loading.py
@@ -29,7 +29,7 @@ from sglang.test.lora_utils import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=45, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=48, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=75, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/lora/test_marlin_lora_correctness.py b/test/registered/lora/test_marlin_lora_correctness.py
index 97a3d71e4..44fc970db 100644
--- a/test/registered/lora/test_marlin_lora_correctness.py
+++ b/test/registered/lora/test_marlin_lora_correctness.py
@@ -34,7 +34,7 @@ from sglang.srt.layers.moe.utils import MoeRunnerBackend
 from sglang.srt.lora.lora_moe_runners import LoRAInfo
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=600, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=129, suite="stage-b-test-1-gpu-large")
 
 
 # ---------------------------------------------------------------------------
diff --git a/test/registered/lora/test_multi_lora_backend.py b/test/registered/lora/test_multi_lora_backend.py
index ac0156bbd..c13acdc93 100644
--- a/test/registered/lora/test_multi_lora_backend.py
+++ b/test/registered/lora/test_multi_lora_backend.py
@@ -25,7 +25,7 @@ from sglang.test.lora_utils import (
 )
 from sglang.test.test_utils import CustomTestCase, is_in_ci
 
-register_cuda_ci(est_time=88, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=99, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=100, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/lora/test_sgemm_sorted_by_adapter.py b/test/registered/lora/test_sgemm_sorted_by_adapter.py
index a8c787267..d1f7a0baf 100644
--- a/test/registered/lora/test_sgemm_sorted_by_adapter.py
+++ b/test/registered/lora/test_sgemm_sorted_by_adapter.py
@@ -7,7 +7,7 @@ import torch
 
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=15, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-large")
 
 
 def _make_batch_info(
diff --git a/test/registered/mla/test_flashmla.py b/test/registered/mla/test_flashmla.py
index 046bcc778..abcad2a39 100644
--- a/test/registered/mla/test_flashmla.py
+++ b/test/registered/mla/test_flashmla.py
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
 )
 
 # FlashMLA attention backend tests with MTP speculative decoding
-register_cuda_ci(est_time=700, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=314, suite="stage-b-test-1-gpu-large")
 
 
 class TestFlashMLAAttnBackend(unittest.TestCase):
diff --git a/test/registered/mla/test_mla.py b/test/registered/mla/test_mla.py
index 0d6761bdb..083d3f5e8 100644
--- a/test/registered/mla/test_mla.py
+++ b/test/registered/mla/test_mla.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
 )
 
 # MLA attention test with MGSM evaluation
-register_cuda_ci(est_time=174, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=181, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=1100, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/mla/test_mla_flashinfer.py b/test/registered/mla/test_mla_flashinfer.py
index 23a24aade..62d269b03 100644
--- a/test/registered/mla/test_mla_flashinfer.py
+++ b/test/registered/mla/test_mla_flashinfer.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
 )
 
 # FlashInfer MLA backend tests with MTP speculative decoding
-register_cuda_ci(est_time=256, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=260, suite="stage-b-test-1-gpu-large")
 
 
 class TestFlashinferMLA(CustomTestCase):
diff --git a/test/registered/mla/test_mla_fp8.py b/test/registered/mla/test_mla_fp8.py
index dd402c5c7..eb793a530 100644
--- a/test/registered/mla/test_mla_fp8.py
+++ b/test/registered/mla/test_mla_fp8.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
 )
 
 # MLA FP8 KV cache test with MGSM evaluation
-register_cuda_ci(est_time=106, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=104, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=800, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/mla/test_mla_int8_deepseek_v3.py b/test/registered/mla/test_mla_int8_deepseek_v3.py
index 35cf4b229..8a94544c8 100644
--- a/test/registered/mla/test_mla_int8_deepseek_v3.py
+++ b/test/registered/mla/test_mla_int8_deepseek_v3.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
 )
 
 # DeepSeek-V3 INT8 quantization tests (channel and block INT8)
-register_cuda_ci(est_time=282, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=313, suite="stage-b-test-1-gpu-large")
 
 
 class TestMLADeepseekV3ChannelInt8(CustomTestCase):
diff --git a/test/registered/model_loading/test_external_models.py b/test/registered/model_loading/test_external_models.py
index d32adfcc9..cd546d998 100644
--- a/test/registered/model_loading/test_external_models.py
+++ b/test/registered/model_loading/test_external_models.py
@@ -5,7 +5,7 @@ from sglang.srt.environ import envs
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=28, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=29, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=45, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/models/test_compressed_tensors_models.py b/test/registered/models/test_compressed_tensors_models.py
index 2d191f40e..74cbe0061 100644
--- a/test/registered/models/test_compressed_tensors_models.py
+++ b/test/registered/models/test_compressed_tensors_models.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=63, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=65, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=42, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/models/test_generation_models.py b/test/registered/models/test_generation_models.py
index 116fc62dc..64d5246f1 100644
--- a/test/registered/models/test_generation_models.py
+++ b/test/registered/models/test_generation_models.py
@@ -1,7 +1,7 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Generation model tests (CUDA only)
-register_cuda_ci(est_time=124, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=150, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=106, suite="stage-b-test-1-gpu-small-amd")
 
 # Copyright 2023-2024 SGLang Team
diff --git a/test/registered/models/test_kimi_linear_models.py b/test/registered/models/test_kimi_linear_models.py
index 83f01d009..21acf3c57 100644
--- a/test/registered/models/test_kimi_linear_models.py
+++ b/test/registered/models/test_kimi_linear_models.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=168, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=178, suite="stage-b-test-2-gpu-large")
 
 
 class TestKimiLinear(CustomTestCase):
diff --git a/test/registered/models/test_nvidia_nemotron_nano_v2.py b/test/registered/models/test_nvidia_nemotron_nano_v2.py
index d9e5883c5..1fb7b67a8 100644
--- a/test/registered/models/test_nvidia_nemotron_nano_v2.py
+++ b/test/registered/models/test_nvidia_nemotron_nano_v2.py
@@ -5,7 +5,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.eval_accuracy_kit import GSM8KMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=214, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=249, suite="stage-b-test-2-gpu-large")
 
 
 class TestNvidiaNemotronNanoV2BF16(GSM8KMixin, DefaultServerBase):
diff --git a/test/registered/models/test_nvidia_nemotron_nano_v2_vl.py b/test/registered/models/test_nvidia_nemotron_nano_v2_vl.py
index 4c52c05d6..510883d8d 100644
--- a/test/registered/models/test_nvidia_nemotron_nano_v2_vl.py
+++ b/test/registered/models/test_nvidia_nemotron_nano_v2_vl.py
@@ -10,7 +10,7 @@ from sglang.test.server_fixtures.mmmu_fixture import MMMUServerBase
 # GSM8k + MMMU evaluation
 
 
-register_cuda_ci(est_time=206, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=256, suite="stage-b-test-1-gpu-large")
 
 MODEL = "nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16"
 
diff --git a/test/registered/models/test_qwen_models.py b/test/registered/models/test_qwen_models.py
index 602de7f25..4283b2809 100644
--- a/test/registered/models/test_qwen_models.py
+++ b/test/registered/models/test_qwen_models.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=109, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=108, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=130, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/models/test_transformers_backend_eval.py b/test/registered/models/test_transformers_backend_eval.py
index 9d68505fa..58698dc5d 100644
--- a/test/registered/models/test_transformers_backend_eval.py
+++ b/test/registered/models/test_transformers_backend_eval.py
@@ -7,7 +7,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=44, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=48, suite="stage-b-test-1-gpu-small")
 
 
 class TestTransformersBackendEval(DefaultServerBase):
diff --git a/test/registered/models/test_transformers_models.py b/test/registered/models/test_transformers_models.py
index 9313bc5ba..5fe40e122 100644
--- a/test/registered/models/test_transformers_models.py
+++ b/test/registered/models/test_transformers_models.py
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=244, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=177, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=320, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/models/test_vlm_models.py b/test/registered/models/test_vlm_models.py
index 2ed573417..17e9b146a 100644
--- a/test/registered/models/test_vlm_models.py
+++ b/test/registered/models/test_vlm_models.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import is_in_ci
 # VLM (Vision Language Model) tests
 
 
-register_cuda_ci(est_time=176, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=317, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=850, suite="stage-b-test-1-gpu-small-amd-nondeterministic")
 
 _is_hip = is_hip()
diff --git a/test/registered/moe/test_cutedsl_moe.py b/test/registered/moe/test_cutedsl_moe.py
index f64db08f0..b0c10db79 100644
--- a/test/registered/moe/test_cutedsl_moe.py
+++ b/test/registered/moe/test_cutedsl_moe.py
@@ -16,7 +16,7 @@ except ImportError:
     CuteDslMoEWrapper = None
     convert_sf_to_mma_layout = None
 
-register_cuda_ci(est_time=590, suite="stage-c-test-4-gpu-b200-small")
+register_cuda_ci(est_time=427, suite="stage-c-test-4-gpu-b200-small")
 
 SKIP_TEST = torch.cuda.get_device_capability() < (10, 0)
 SKIP_REASON = "Nvfp4 Requires compute capability of 10 or above."
diff --git a/test/registered/moe/test_fused_moe.py b/test/registered/moe/test_fused_moe.py
index b67d1b4ed..f93cbff99 100644
--- a/test/registered/moe/test_fused_moe.py
+++ b/test/registered/moe/test_fused_moe.py
@@ -13,7 +13,7 @@ from sglang.srt.utils import get_device, get_device_capability, is_hip
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase, empty_gpu_cache
 
-register_cuda_ci(est_time=79, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=87, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=30, suite="stage-b-test-1-gpu-small-amd")
 
 _is_hip = is_hip()
diff --git a/test/registered/moe/test_glm4_moe_models.py b/test/registered/moe/test_glm4_moe_models.py
index 58a54f25b..e6e800ecd 100644
--- a/test/registered/moe/test_glm4_moe_models.py
+++ b/test/registered/moe/test_glm4_moe_models.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=150, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=171, suite="stage-b-test-2-gpu-large")
 
 
 class TestGLM4MoE(CustomTestCase):
diff --git a/test/registered/moe/test_moe_ep.py b/test/registered/moe/test_moe_ep.py
index 5aa4d4836..b5e936ee6 100644
--- a/test/registered/moe/test_moe_ep.py
+++ b/test/registered/moe/test_moe_ep.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=232, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=279, suite="stage-b-test-2-gpu-large")
 
 
 class TestEp(CustomTestCase):
diff --git a/test/registered/moe/test_torch_compile_moe.py b/test/registered/moe/test_torch_compile_moe.py
index 2175f0dc9..811ca69b4 100644
--- a/test/registered/moe/test_torch_compile_moe.py
+++ b/test/registered/moe/test_torch_compile_moe.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=125, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=130, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=1400, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/moe/test_triton_fused_moe.py b/test/registered/moe/test_triton_fused_moe.py
index b5c263a04..309ac1b3a 100644
--- a/test/registered/moe/test_triton_fused_moe.py
+++ b/test/registered/moe/test_triton_fused_moe.py
@@ -12,7 +12,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=12, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=13, suite="stage-b-test-1-gpu-large")
 
 
 class TestFusedMOE(CustomTestCase):
diff --git a/test/registered/moe/test_triton_moe_channel_fp8_kernel.py b/test/registered/moe/test_triton_moe_channel_fp8_kernel.py
index 49bdff402..ee6b45227 100644
--- a/test/registered/moe/test_triton_moe_channel_fp8_kernel.py
+++ b/test/registered/moe/test_triton_moe_channel_fp8_kernel.py
@@ -11,7 +11,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=16, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=17, suite="stage-b-test-1-gpu-large")
 
 
 def native_w8a8_per_token_matmul(A, B, As, Bs, output_dtype=torch.float16):
diff --git a/test/registered/observability/test_metrics.py b/test/registered/observability/test_metrics.py
index 066999d02..2b6c65987 100644
--- a/test/registered/observability/test_metrics.py
+++ b/test/registered/observability/test_metrics.py
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=73, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=74, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=32, suite="stage-b-test-1-gpu-small-amd")
 
 _MODEL_NAME = "Qwen/Qwen3-0.6B"
diff --git a/test/registered/observability/test_tracing.py b/test/registered/observability/test_tracing.py
index f77dc2ea2..4217a1ac0 100644
--- a/test/registered/observability/test_tracing.py
+++ b/test/registered/observability/test_tracing.py
@@ -46,7 +46,7 @@ from sglang.test.test_utils import (
 logger = logging.getLogger(__name__)
 
 # CI registration
-register_cuda_ci(est_time=104, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=113, suite="stage-b-test-1-gpu-small")
 
 
 # ============================================================================
diff --git a/test/registered/observability/test_tracing_disaggregation.py b/test/registered/observability/test_tracing_disaggregation.py
index 91d1e11ed..3c7990df7 100644
--- a/test/registered/observability/test_tracing_disaggregation.py
+++ b/test/registered/observability/test_tracing_disaggregation.py
@@ -33,7 +33,7 @@ from sglang.utils import wait_for_http_ready
 logger = logging.getLogger(__name__)
 
 # CI registration - PD disaggregation requires 2 GPUs
-register_cuda_ci(est_time=48, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=65, suite="stage-b-test-2-gpu-large")
 
 
 class TestTraceDisaggregation(CustomTestCase):
diff --git a/test/registered/openai_server/basic/test_anthropic_server.py b/test/registered/openai_server/basic/test_anthropic_server.py
index 3e50e6758..bdfb8b91f 100644
--- a/test/registered/openai_server/basic/test_anthropic_server.py
+++ b/test/registered/openai_server/basic/test_anthropic_server.py
@@ -31,7 +31,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=39, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=40, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=140, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/openai_server/basic/test_http2_server.py b/test/registered/openai_server/basic/test_http2_server.py
index d35204c31..e2a05a335 100644
--- a/test/registered/openai_server/basic/test_http2_server.py
+++ b/test/registered/openai_server/basic/test_http2_server.py
@@ -27,7 +27,7 @@ try:
 except ImportError:
     _HAS_GRANIAN = False
 
-register_cuda_ci(est_time=51, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=52, suite="stage-b-test-1-gpu-small")
 
 
 @unittest.skipUnless(_HAS_GRANIAN, "granian not installed (pip install sglang[http2])")
diff --git a/test/registered/openai_server/basic/test_openai_server.py b/test/registered/openai_server/basic/test_openai_server.py
index 967ef109b..4a96c100f 100644
--- a/test/registered/openai_server/basic/test_openai_server.py
+++ b/test/registered/openai_server/basic/test_openai_server.py
@@ -28,7 +28,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=189, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=182, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=200, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/openai_server/basic/test_serving_transcription.py b/test/registered/openai_server/basic/test_serving_transcription.py
index 6a51d36a7..31efdb5be 100644
--- a/test/registered/openai_server/basic/test_serving_transcription.py
+++ b/test/registered/openai_server/basic/test_serving_transcription.py
@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=51, suite="stage-b-test-1-gpu-small")
 
 WHISPER_MODEL = "openai/whisper-large-v3"
 AUDIO_URL = "https://raw.githubusercontent.com/sgl-project/sgl-test-files/refs/heads/main/audios/Trump_WEF_2018_10s.mp3"
diff --git a/test/registered/openai_server/features/test_json_mode.py b/test/registered/openai_server/features/test_json_mode.py
index fcdd6e792..d3aa8e68a 100644
--- a/test/registered/openai_server/features/test_json_mode.py
+++ b/test/registered/openai_server/features/test_json_mode.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=108, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=118, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=180, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/openai_server/features/test_openai_server_ebnf.py b/test/registered/openai_server/features/test_openai_server_ebnf.py
index 78327bc50..42dfd58fd 100644
--- a/test/registered/openai_server/features/test_openai_server_ebnf.py
+++ b/test/registered/openai_server/features/test_openai_server_ebnf.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=42, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=44, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/openai_server/features/test_openai_server_hidden_states.py b/test/registered/openai_server/features/test_openai_server_hidden_states.py
index 868e9aa7a..c1a46fd7a 100644
--- a/test/registered/openai_server/features/test_openai_server_hidden_states.py
+++ b/test/registered/openai_server/features/test_openai_server_hidden_states.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=193, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=222, suite="stage-b-test-1-gpu-small")
 register_amd_ci(
     est_time=186,
     suite="stage-b-test-1-gpu-small-amd",
diff --git a/test/registered/openai_server/function_call/test_openai_function_calling.py b/test/registered/openai_server/function_call/test_openai_function_calling.py
index fc3847b6c..88544c197 100644
--- a/test/registered/openai_server/function_call/test_openai_function_calling.py
+++ b/test/registered/openai_server/function_call/test_openai_function_calling.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=97, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=100, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=73, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/openai_server/function_call/test_tool_choice.py b/test/registered/openai_server/function_call/test_tool_choice.py
index 1c7e49617..c7fef05c7 100644
--- a/test/registered/openai_server/function_call/test_tool_choice.py
+++ b/test/registered/openai_server/function_call/test_tool_choice.py
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=194, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=204, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=258, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/openai_server/validation/test_large_max_new_tokens.py b/test/registered/openai_server/validation/test_large_max_new_tokens.py
index c48ca4547..3ec576774 100644
--- a/test/registered/openai_server/validation/test_large_max_new_tokens.py
+++ b/test/registered/openai_server/validation/test_large_max_new_tokens.py
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=56, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=58, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=41, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/openai_server/validation/test_openai_server_ignore_eos.py b/test/registered/openai_server/validation/test_openai_server_ignore_eos.py
index 2986c2f57..847dd0463 100644
--- a/test/registered/openai_server/validation/test_openai_server_ignore_eos.py
+++ b/test/registered/openai_server/validation/test_openai_server_ignore_eos.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=42, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=44, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=47, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/openai_server/validation/test_request_length_validation.py b/test/registered/openai_server/validation/test_request_length_validation.py
index 1bc93e551..0c581bed0 100644
--- a/test/registered/openai_server/validation/test_request_length_validation.py
+++ b/test/registered/openai_server/validation/test_request_length_validation.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=39, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=49, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=31, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/perf/test_bench_one_batch_1gpu.py b/test/registered/perf/test_bench_one_batch_1gpu.py
index 26c251ccd..13f16dc92 100644
--- a/test/registered/perf/test_bench_one_batch_1gpu.py
+++ b/test/registered/perf/test_bench_one_batch_1gpu.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=80, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=95, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=120, suite="stage-b-test-1-gpu-large-amd")
 
 
diff --git a/test/registered/perf/test_bench_one_batch_2gpu.py b/test/registered/perf/test_bench_one_batch_2gpu.py
index 41ba7a030..2daa9b6c4 100644
--- a/test/registered/perf/test_bench_one_batch_2gpu.py
+++ b/test/registered/perf/test_bench_one_batch_2gpu.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=167, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=209, suite="stage-b-test-2-gpu-large")
 register_amd_ci(est_time=630, suite="stage-b-test-2-gpu-large-amd")
 
 
diff --git a/test/registered/perf/test_bench_serving_1gpu_large.py b/test/registered/perf/test_bench_serving_1gpu_large.py
index 43e0ad63e..cee6d4014 100644
--- a/test/registered/perf/test_bench_serving_1gpu_large.py
+++ b/test/registered/perf/test_bench_serving_1gpu_large.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=251, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=286, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=300, suite="stage-b-test-1-gpu-large-amd")
 
 
diff --git a/test/registered/perf/test_bench_serving_1gpu_part1.py b/test/registered/perf/test_bench_serving_1gpu_part1.py
index 635593307..56040ffb6 100644
--- a/test/registered/perf/test_bench_serving_1gpu_part1.py
+++ b/test/registered/perf/test_bench_serving_1gpu_part1.py
@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=1184, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=1210, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=1100, suite="stage-b-test-1-gpu-large-amd")
 
 
diff --git a/test/registered/perf/test_bench_serving_1gpu_part2.py b/test/registered/perf/test_bench_serving_1gpu_part2.py
index 23eb51a22..7743877b0 100644
--- a/test/registered/perf/test_bench_serving_1gpu_part2.py
+++ b/test/registered/perf/test_bench_serving_1gpu_part2.py
@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=910, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=968, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=900, suite="stage-b-test-1-gpu-large-amd")
 
 
diff --git a/test/registered/perf/test_bench_serving_2gpu.py b/test/registered/perf/test_bench_serving_2gpu.py
index d03cf2499..91783b07b 100644
--- a/test/registered/perf/test_bench_serving_2gpu.py
+++ b/test/registered/perf/test_bench_serving_2gpu.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=660, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=721, suite="stage-b-test-2-gpu-large")
 register_amd_ci(est_time=1450, suite="stage-b-test-2-gpu-large-amd")
 
 
diff --git a/test/registered/perf/test_vlm_perf_5090.py b/test/registered/perf/test_vlm_perf_5090.py
index 8802c75b3..83f1ce25a 100644
--- a/test/registered/perf/test_vlm_perf_5090.py
+++ b/test/registered/perf/test_vlm_perf_5090.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=410, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=406, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=500, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/piecewise_cuda_graph/test_pcg_with_speculative_decoding.py b/test/registered/piecewise_cuda_graph/test_pcg_with_speculative_decoding.py
index e65f16b3f..db8eea094 100644
--- a/test/registered/piecewise_cuda_graph/test_pcg_with_speculative_decoding.py
+++ b/test/registered/piecewise_cuda_graph/test_pcg_with_speculative_decoding.py
@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=600, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=531, suite="stage-b-test-2-gpu-large")
 
 
 class TestPCGWithMTP(unittest.TestCase):
diff --git a/test/registered/piecewise_cuda_graph/test_piecewise_cuda_graph_support_1_gpu.py b/test/registered/piecewise_cuda_graph/test_piecewise_cuda_graph_support_1_gpu.py
index 9d4a785a9..56e12fbd0 100644
--- a/test/registered/piecewise_cuda_graph/test_piecewise_cuda_graph_support_1_gpu.py
+++ b/test/registered/piecewise_cuda_graph/test_piecewise_cuda_graph_support_1_gpu.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
 )
 
 # CI Registration
-register_cuda_ci(est_time=262, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=260, suite="stage-b-test-1-gpu-large")
 
 
 class TestPiecewiseCudaGraphQwen25VL(CustomTestCase):
diff --git a/test/registered/prefill_only/test_cross_encoder_models.py b/test/registered/prefill_only/test_cross_encoder_models.py
index d63323d54..c32c7eac5 100644
--- a/test/registered/prefill_only/test_cross_encoder_models.py
+++ b/test/registered/prefill_only/test_cross_encoder_models.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import CustomTestCase, is_in_ci
 # Cross encoder model tests
 
 
-register_cuda_ci(est_time=101, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=125, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=150, suite="stage-b-test-1-gpu-small-amd")
 
 MODELS = [
diff --git a/test/registered/prefill_only/test_embed_overrides.py b/test/registered/prefill_only/test_embed_overrides.py
index 01349dbd1..f0c8ff830 100644
--- a/test/registered/prefill_only/test_embed_overrides.py
+++ b/test/registered/prefill_only/test_embed_overrides.py
@@ -24,7 +24,7 @@ from sglang.srt.server_args import MIS_DELIMITER_TOKEN_ID
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
 
 HIDDEN_DIM = 4
 
diff --git a/test/registered/prefill_only/test_embedding_models.py b/test/registered/prefill_only/test_embedding_models.py
index 6776c8602..7dc4aa6c8 100644
--- a/test/registered/prefill_only/test_embedding_models.py
+++ b/test/registered/prefill_only/test_embedding_models.py
@@ -35,7 +35,7 @@ register_amd_ci(
     suite="stage-b-test-1-gpu-small-amd",
     disabled="see https://github.com/sgl-project/sglang/issues/11127",
 )
-register_cuda_ci(est_time=102, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=136, suite="stage-b-test-1-gpu-small")
 
 MODEL_TO_CONFIG = {
     "Alibaba-NLP/gte-Qwen2-1.5B-instruct": (1, 1e-5),
diff --git a/test/registered/prefill_only/test_encoder_embedding_models.py b/test/registered/prefill_only/test_encoder_embedding_models.py
index 90cfca661..8f9a694a9 100644
--- a/test/registered/prefill_only/test_encoder_embedding_models.py
+++ b/test/registered/prefill_only/test_encoder_embedding_models.py
@@ -29,7 +29,7 @@ from sglang.test.test_utils import CustomTestCase, get_similarities, is_in_ci
 # python -m unittest test_encoder_embedding_models.TestEncoderEmbeddingModels.test_prefill_logits
 
 
-register_cuda_ci(est_time=198, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=444, suite="stage-b-test-1-gpu-small")
 
 MODELS = [("BAAI/bge-small-en", 1, 1e-5), ("BAAI/bge-m3", 1, 1e-5)]
 
diff --git a/test/registered/prefill_only/test_multi_item_scoring.py b/test/registered/prefill_only/test_multi_item_scoring.py
index 7ec80e98c..fb9205cdd 100644
--- a/test/registered/prefill_only/test_multi_item_scoring.py
+++ b/test/registered/prefill_only/test_multi_item_scoring.py
@@ -27,7 +27,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
-register_cuda_ci(est_time=240, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=211, suite="stage-b-test-1-gpu-small")
 
 TEST_MODEL_NAME = os.environ.get("TEST_MODEL_NAME", DEFAULT_SMALL_MODEL_NAME_FOR_TEST)
 TEST_CLASSIFICATION_BASE_MODEL = os.environ.get(
diff --git a/test/registered/prefill_only/test_openai_embedding.py b/test/registered/prefill_only/test_openai_embedding.py
index cb3507027..cb088aa17 100644
--- a/test/registered/prefill_only/test_openai_embedding.py
+++ b/test/registered/prefill_only/test_openai_embedding.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=80, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=91, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=141, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/prefill_only/test_pooled_hidden_states.py b/test/registered/prefill_only/test_pooled_hidden_states.py
index dd1757302..0d3b7b969 100644
--- a/test/registered/prefill_only/test_pooled_hidden_states.py
+++ b/test/registered/prefill_only/test_pooled_hidden_states.py
@@ -27,7 +27,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=240, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=100, suite="stage-b-test-1-gpu-small")
 
 _SEQCLS_MODEL = "Qwen/Qwen3-0.6B"
 _CAUSAL_LM_MODEL = DEFAULT_SMALL_MODEL_NAME_FOR_TEST
diff --git a/test/registered/prefill_only/test_reward_models.py b/test/registered/prefill_only/test_reward_models.py
index fb3084675..53eda08bb 100644
--- a/test/registered/prefill_only/test_reward_models.py
+++ b/test/registered/prefill_only/test_reward_models.py
@@ -24,7 +24,7 @@ from sglang.test.test_utils import CustomTestCase
 # ==============================================================================
 
 
-register_cuda_ci(est_time=142, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=166, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=132, suite="stage-b-test-1-gpu-small-amd-nondeterministic")
 
 MODELS = [
diff --git a/test/registered/prefill_only/test_score_api.py b/test/registered/prefill_only/test_score_api.py
index 4def12765..ae06a5835 100644
--- a/test/registered/prefill_only/test_score_api.py
+++ b/test/registered/prefill_only/test_score_api.py
@@ -27,7 +27,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=70, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=71, suite="stage-b-test-1-gpu-small")
 
 _MODEL = os.environ.get("TEST_MODEL_NAME", DEFAULT_SMALL_MODEL_NAME_FOR_TEST)
 
diff --git a/test/registered/prefill_only/test_serving_rerank.py b/test/registered/prefill_only/test_serving_rerank.py
index 411ea1c8e..f1b10f253 100644
--- a/test/registered/prefill_only/test_serving_rerank.py
+++ b/test/registered/prefill_only/test_serving_rerank.py
@@ -7,7 +7,7 @@ from sglang.srt.managers.tokenizer_manager_score_mixin import ScoreResult
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # Keep consistent with other openai_server/basic unit tests.
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=10, suite="stage-b-test-1-gpu-small-amd")
 
 try:
diff --git a/test/registered/profiling/test_start_profile.py b/test/registered/profiling/test_start_profile.py
index edfb48fa5..9049f9cba 100644
--- a/test/registered/profiling/test_start_profile.py
+++ b/test/registered/profiling/test_start_profile.py
@@ -29,7 +29,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=40, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=42, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=60, suite="stage-b-test-1-gpu-small-amd")
 
 OUTPUT_DIR = "./profiler_dir"
diff --git a/test/registered/quant/test_autoround.py b/test/registered/quant/test_autoround.py
index 9c1e3005f..f71967b18 100644
--- a/test/registered/quant/test_autoround.py
+++ b/test/registered/quant/test_autoround.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=96, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=99, suite="stage-b-test-1-gpu-large")
 
 
 class TestAutoRound(CustomTestCase):
diff --git a/test/registered/quant/test_awq.py b/test/registered/quant/test_awq.py
index 0f72d8ded..ae13e8431 100644
--- a/test/registered/quant/test_awq.py
+++ b/test/registered/quant/test_awq.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=950, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=226, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=200, suite="stage-b-test-1-gpu-large-amd")
 
 
diff --git a/test/registered/quant/test_block_int8.py b/test/registered/quant/test_block_int8.py
index aa0c8bae5..b7716c8dc 100644
--- a/test/registered/quant/test_block_int8.py
+++ b/test/registered/quant/test_block_int8.py
@@ -10,7 +10,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=39, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=44, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=22, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/quant/test_deepseek_v32_fp4_4gpu.py b/test/registered/quant/test_deepseek_v32_fp4_4gpu.py
index dc5d2824b..9860c38c5 100644
--- a/test/registered/quant/test_deepseek_v32_fp4_4gpu.py
+++ b/test/registered/quant/test_deepseek_v32_fp4_4gpu.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=852, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=874, suite="stage-c-test-4-gpu-b200")
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3.2-NVFP4"
 SERVER_LAUNCH_TIMEOUT = 1200
diff --git a/test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py b/test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py
index 8ea65b8ca..59dbe8741 100644
--- a/test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py
+++ b/test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=1030, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=1060, suite="stage-c-test-4-gpu-b200")
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3.2-NVFP4"
 SERVER_LAUNCH_TIMEOUT = 1200
diff --git a/test/registered/quant/test_deepseek_v3_fp4_4gpu.py b/test/registered/quant/test_deepseek_v3_fp4_4gpu.py
index e667bbc6f..5ce8a8fb3 100644
--- a/test/registered/quant/test_deepseek_v3_fp4_4gpu.py
+++ b/test/registered/quant/test_deepseek_v3_fp4_4gpu.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=1380, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=1190, suite="stage-c-test-4-gpu-b200")
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
 SERVER_LAUNCH_TIMEOUT = 1200
diff --git a/test/registered/quant/test_eval_fp8_accuracy.py b/test/registered/quant/test_eval_fp8_accuracy.py
index 5fa425a24..e91e683c0 100644
--- a/test/registered/quant/test_eval_fp8_accuracy.py
+++ b/test/registered/quant/test_eval_fp8_accuracy.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=326, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=351, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=600, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/quant/test_fp8_gemm_sm120.py b/test/registered/quant/test_fp8_gemm_sm120.py
index bbdf97fa6..9ecbb98ec 100644
--- a/test/registered/quant/test_fp8_gemm_sm120.py
+++ b/test/registered/quant/test_fp8_gemm_sm120.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=144, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=146, suite="stage-b-test-1-gpu-small")
 
 PERTENSOR_MODEL_PATH = "nvidia/Llama-3.1-8B-Instruct-FP8"
 BLOCKWISE_MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507-FP8"
diff --git a/test/registered/quant/test_fp8_utils.py b/test/registered/quant/test_fp8_utils.py
index b4118d6f6..563a9123b 100644
--- a/test/registered/quant/test_fp8_utils.py
+++ b/test/registered/quant/test_fp8_utils.py
@@ -10,7 +10,7 @@ from sglang.srt.layers.quantization.fp8_utils import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-large")
 
 
 class TestInverseTransformScaleUe8m0(CustomTestCase):
diff --git a/test/registered/quant/test_fp8kv_triton.py b/test/registered/quant/test_fp8kv_triton.py
index 8136f0109..c46302444 100644
--- a/test/registered/quant/test_fp8kv_triton.py
+++ b/test/registered/quant/test_fp8kv_triton.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=63, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=73, suite="stage-b-test-1-gpu-large")
 
 
 class TestFP8KVCacheTritonBackend(CustomTestCase):
diff --git a/test/registered/quant/test_gguf.py b/test/registered/quant/test_gguf.py
index 4ae3bb8e0..3ec75dad0 100644
--- a/test/registered/quant/test_gguf.py
+++ b/test/registered/quant/test_gguf.py
@@ -6,7 +6,7 @@ import sglang as sgl
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=71, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=76, suite="stage-b-test-1-gpu-small")
 
 
 class TestGGUF(CustomTestCase):
diff --git a/test/registered/quant/test_gptqmodel_dynamic.py b/test/registered/quant/test_gptqmodel_dynamic.py
index 15927a7c9..e0f542aa2 100644
--- a/test/registered/quant/test_gptqmodel_dynamic.py
+++ b/test/registered/quant/test_gptqmodel_dynamic.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=98, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=100, suite="stage-b-test-1-gpu-large")
 
 
 def check_quant_method(model_path: str, use_marlin_kernel: bool):
diff --git a/test/registered/quant/test_int8_kernel.py b/test/registered/quant/test_int8_kernel.py
index 4114887dd..8b48a913a 100644
--- a/test/registered/quant/test_int8_kernel.py
+++ b/test/registered/quant/test_int8_kernel.py
@@ -11,7 +11,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=13, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=15, suite="stage-b-test-1-gpu-small")
 
 
 def native_w8a8_per_token_matmul(A, B, As, Bs, output_dtype=torch.float16):
diff --git a/test/registered/quant/test_marlin_moe.py b/test/registered/quant/test_marlin_moe.py
index 9554c51f6..d88a3627e 100644
--- a/test/registered/quant/test_marlin_moe.py
+++ b/test/registered/quant/test_marlin_moe.py
@@ -12,7 +12,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_marlin_utils import awq_marlin_quantize, marlin_quantize
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=85, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=108, suite="stage-b-test-1-gpu-small")
 
 set_global_server_args_for_scheduler(object.__new__(ServerArgs))
 
diff --git a/test/registered/quant/test_nvfp4_gemm_sm120.py b/test/registered/quant/test_nvfp4_gemm_sm120.py
index 4f87b965b..7b1f4390f 100644
--- a/test/registered/quant/test_nvfp4_gemm_sm120.py
+++ b/test/registered/quant/test_nvfp4_gemm_sm120.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=115, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=109, suite="stage-b-test-1-gpu-small")
 
 MODEL_PATH = "nvidia/Llama-3.1-8B-Instruct-NVFP4"
 
diff --git a/test/registered/quant/test_quant_config_parsing.py b/test/registered/quant/test_quant_config_parsing.py
index 5db3be7b8..33ebda5b8 100644
--- a/test/registered/quant/test_quant_config_parsing.py
+++ b/test/registered/quant/test_quant_config_parsing.py
@@ -5,7 +5,7 @@ from sglang.srt.configs.model_config import ModelConfig
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=12, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=15, suite="stage-a-test-cpu")
 
 
 class TestQuantLogString(CustomTestCase):
diff --git a/test/registered/quant/test_quantization.py b/test/registered/quant/test_quantization.py
index 30ccc18f8..8bf8401a4 100644
--- a/test/registered/quant/test_quantization.py
+++ b/test/registered/quant/test_quantization.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     write_results_to_json,
 )
 
-register_cuda_ci(est_time=402, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=460, suite="stage-b-test-1-gpu-large")
 
 MODEL_SCORE_THRESHOLDS = {
     # Baselines observed with gsm8k 5-shot concatenated format via chat API,
diff --git a/test/registered/quant/test_triton_scaled_mm.py b/test/registered/quant/test_triton_scaled_mm.py
index f66c33c5d..8057fc593 100644
--- a/test/registered/quant/test_triton_scaled_mm.py
+++ b/test/registered/quant/test_triton_scaled_mm.py
@@ -8,7 +8,7 @@ from sglang.srt.layers.quantization.fp8_kernel import triton_scaled_mm
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=12, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/quant/test_w4a8_deepseek_v3.py b/test/registered/quant/test_w4a8_deepseek_v3.py
index 72e717eb8..622c9e3c8 100644
--- a/test/registered/quant/test_w4a8_deepseek_v3.py
+++ b/test/registered/quant/test_w4a8_deepseek_v3.py
@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=700, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=930, suite="stage-c-test-8-gpu-h20")
 
 
 class TestDeepseekV3W4afp8(CustomTestCase):
diff --git a/test/registered/quant/test_w8a8_quantization.py b/test/registered/quant/test_w8a8_quantization.py
index 33805704e..7841d098d 100644
--- a/test/registered/quant/test_w8a8_quantization.py
+++ b/test/registered/quant/test_w8a8_quantization.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=204, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=232, suite="stage-b-test-1-gpu-large")
 
 
 class BaseW8A8Test(CustomTestCase):
diff --git a/test/registered/radix_cache/test_radix_attention.py b/test/registered/radix_cache/test_radix_attention.py
index 3ce4643da..871dac26e 100644
--- a/test/registered/radix_cache/test_radix_attention.py
+++ b/test/registered/radix_cache/test_radix_attention.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
 )
 
 # RadixAttention server integration tests
-register_cuda_ci(est_time=96, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=100, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=100, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/radix_cache/test_radix_cache_hit.py b/test/registered/radix_cache/test_radix_cache_hit.py
index 974adb866..18bd3a997 100644
--- a/test/registered/radix_cache/test_radix_cache_hit.py
+++ b/test/registered/radix_cache/test_radix_cache_hit.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=62, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=55, suite="stage-b-test-1-gpu-small")
 
 MODEL = DEFAULT_SMALL_MODEL_NAME_FOR_TEST
 
diff --git a/test/registered/radix_cache/test_swa_radix_cache_kl.py b/test/registered/radix_cache/test_swa_radix_cache_kl.py
index a58abd916..02a336384 100644
--- a/test/registered/radix_cache/test_swa_radix_cache_kl.py
+++ b/test/registered/radix_cache/test_swa_radix_cache_kl.py
@@ -6,7 +6,7 @@ from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
 MODEL = "openai/gpt-oss-20b"
 
-register_cuda_ci(est_time=142, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=151, suite="stage-b-test-1-gpu-large")
 
 
 class TestSWARadixCacheKL(KLDivergenceMixin, DefaultServerBase):
diff --git a/test/registered/radix_cache/test_unified_radix_cache_kl.py b/test/registered/radix_cache/test_unified_radix_cache_kl.py
index abf1324c3..6ac338518 100644
--- a/test/registered/radix_cache/test_unified_radix_cache_kl.py
+++ b/test/registered/radix_cache/test_unified_radix_cache_kl.py
@@ -34,7 +34,7 @@ MAMBA_TRACK_INTERVAL = 128
 SWA_MODEL = "openai/gpt-oss-20b"
 FULL_MODEL = "Qwen/Qwen3-32B"
 
-register_cuda_ci(est_time=1200, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=632, suite="stage-c-test-4-gpu-h100")
 
 
 class UnifiedRadixTreeTestMixin:
diff --git a/test/registered/reasoning/test_reasoning.py b/test/registered/reasoning/test_reasoning.py
index b6295c3da..b271bea52 100644
--- a/test/registered/reasoning/test_reasoning.py
+++ b/test/registered/reasoning/test_reasoning.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=111, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=129, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=200, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/rl/test_fp32_lm_head.py b/test/registered/rl/test_fp32_lm_head.py
index f98769cdb..8b721cc4e 100644
--- a/test/registered/rl/test_fp32_lm_head.py
+++ b/test/registered/rl/test_fp32_lm_head.py
@@ -15,7 +15,7 @@ from sglang.srt.server_args import (
 from sglang.srt.utils import get_device
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=15, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/rl/test_lora_load_from_tensor.py b/test/registered/rl/test_lora_load_from_tensor.py
index 14bb64c7d..6dd6a89bd 100644
--- a/test/registered/rl/test_lora_load_from_tensor.py
+++ b/test/registered/rl/test_lora_load_from_tensor.py
@@ -10,7 +10,7 @@ import sglang as sgl
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=78, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=102, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=90, suite="stage-b-test-1-gpu-small-amd")
 
 MODEL_PATH = "Qwen/Qwen3-0.6B"
diff --git a/test/registered/rl/test_patch_torch.py b/test/registered/rl/test_patch_torch.py
index 445713165..260ae2f30 100644
--- a/test/registered/rl/test_patch_torch.py
+++ b/test/registered/rl/test_patch_torch.py
@@ -12,7 +12,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 register_amd_ci(
     est_time=19, suite="stage-b-test-2-gpu-large-amd", disabled="see #11127"
 )
-register_cuda_ci(est_time=22, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=38, suite="stage-b-test-2-gpu-large")
 
 
 class TestReleaseMemoryOccupation(unittest.TestCase):
diff --git a/test/registered/rl/test_pause_generation_tensor_consistency.py b/test/registered/rl/test_pause_generation_tensor_consistency.py
index 0223cca11..64c172221 100644
--- a/test/registered/rl/test_pause_generation_tensor_consistency.py
+++ b/test/registered/rl/test_pause_generation_tensor_consistency.py
@@ -9,7 +9,7 @@ import torch
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 # ---------------------------------------------------------------------------
diff --git a/test/registered/rl/test_return_routed_experts.py b/test/registered/rl/test_return_routed_experts.py
index d75bbd4e2..a4b4f6407 100644
--- a/test/registered/rl/test_return_routed_experts.py
+++ b/test/registered/rl/test_return_routed_experts.py
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=164, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=185, suite="stage-b-test-2-gpu-large")
 register_amd_ci(
     est_time=200,
     suite="stage-b-test-2-gpu-large-amd",
diff --git a/test/registered/rl/test_update_weights_from_disk_blackwell.py b/test/registered/rl/test_update_weights_from_disk_blackwell.py
index 935d8ba31..ba13c5bbd 100644
--- a/test/registered/rl/test_update_weights_from_disk_blackwell.py
+++ b/test/registered/rl/test_update_weights_from_disk_blackwell.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=149, suite="stage-c-test-4-gpu-b200-small")
+register_cuda_ci(est_time=400, suite="stage-c-test-4-gpu-b200-small")
 
 import unittest
 
diff --git a/test/registered/rl/test_update_weights_from_distributed.py b/test/registered/rl/test_update_weights_from_distributed.py
index d728fbc53..321826cca 100644
--- a/test/registered/rl/test_update_weights_from_distributed.py
+++ b/test/registered/rl/test_update_weights_from_distributed.py
@@ -43,7 +43,7 @@ from sglang.test.test_utils import (
 )
 from sglang.utils import terminate_process
 
-register_cuda_ci(est_time=91, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=137, suite="stage-b-test-2-gpu-large")
 register_amd_ci(est_time=400, suite="stage-b-test-2-gpu-large-amd")
 
 mp.set_start_method("spawn", force=True)
diff --git a/test/registered/rl/test_update_weights_from_tensor.py b/test/registered/rl/test_update_weights_from_tensor.py
index cdd1afdc4..298030975 100644
--- a/test/registered/rl/test_update_weights_from_tensor.py
+++ b/test/registered/rl/test_update_weights_from_tensor.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=136, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=147, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=195, suite="stage-b-test-1-gpu-small-amd")
 
 import gc
diff --git a/test/registered/sampling/test_original_logprobs.py b/test/registered/sampling/test_original_logprobs.py
index b8125e407..c5a21f0e3 100644
--- a/test/registered/sampling/test_original_logprobs.py
+++ b/test/registered/sampling/test_original_logprobs.py
@@ -25,7 +25,7 @@ import sglang as sgl
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST
 
-register_cuda_ci(est_time=46, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=45, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=60, suite="stage-b-test-1-gpu-small-amd")
 
 # ------------------------- Configurable via env ------------------------- #
diff --git a/test/registered/sampling/test_penalty.py b/test/registered/sampling/test_penalty.py
index 06311c083..d6b7425a1 100644
--- a/test/registered/sampling/test_penalty.py
+++ b/test/registered/sampling/test_penalty.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=50, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=53, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=82, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/sampling/test_pytorch_sampling_backend.py b/test/registered/sampling/test_pytorch_sampling_backend.py
index 23e10e05b..ad13ebda8 100644
--- a/test/registered/sampling/test_pytorch_sampling_backend.py
+++ b/test/registered/sampling/test_pytorch_sampling_backend.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=77, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=80, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=66, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/scheduler/test_abort_with_metrics.py b/test/registered/scheduler/test_abort_with_metrics.py
index 3c1a8c0b0..4f894a31f 100644
--- a/test/registered/scheduler/test_abort_with_metrics.py
+++ b/test/registered/scheduler/test_abort_with_metrics.py
@@ -18,7 +18,7 @@ from sglang.srt.utils.http_middleware_patch import _PureASGIDispatch
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 _HTTP_SCOPE = {
     "type": "http",
diff --git a/test/registered/scheduler/test_mixed_chunked_prefill.py b/test/registered/scheduler/test_mixed_chunked_prefill.py
index 895943394..cced856d3 100644
--- a/test/registered/scheduler/test_mixed_chunked_prefill.py
+++ b/test/registered/scheduler/test_mixed_chunked_prefill.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=180, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=167, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=180, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/scheduler/test_no_chunked_prefill.py b/test/registered/scheduler/test_no_chunked_prefill.py
index 5aa67fc14..db574e37b 100644
--- a/test/registered/scheduler/test_no_chunked_prefill.py
+++ b/test/registered/scheduler/test_no_chunked_prefill.py
@@ -8,7 +8,7 @@ from sglang.test.test_utils import (
     run_mmlu_test,
 )
 
-register_cuda_ci(est_time=121, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=131, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=108, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/scheduler/test_no_overlap_scheduler.py b/test/registered/scheduler/test_no_overlap_scheduler.py
index b3e0b28f6..c3e972892 100644
--- a/test/registered/scheduler/test_no_overlap_scheduler.py
+++ b/test/registered/scheduler/test_no_overlap_scheduler.py
@@ -9,7 +9,7 @@ import unittest
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase, run_mmlu_test
 
-register_cuda_ci(est_time=231, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=267, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=275, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/scheduler/test_priority_scheduling.py b/test/registered/scheduler/test_priority_scheduling.py
index d9328c857..71e9cccb6 100644
--- a/test/registered/scheduler/test_priority_scheduling.py
+++ b/test/registered/scheduler/test_priority_scheduling.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     send_concurrent_generate_requests_with_custom_params,
 )
 
-register_cuda_ci(est_time=133, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=149, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=195, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/scheduler/test_retract_decode.py b/test/registered/scheduler/test_retract_decode.py
index 45e120ca1..56ed5b0d4 100644
--- a/test/registered/scheduler/test_retract_decode.py
+++ b/test/registered/scheduler/test_retract_decode.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
 )
 from sglang.utils import is_in_ci
 
-register_cuda_ci(est_time=305, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=353, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=600, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/scheduler/test_scheduler_control.py b/test/registered/scheduler/test_scheduler_control.py
index 3e6d967d2..ea77c8023 100644
--- a/test/registered/scheduler/test_scheduler_control.py
+++ b/test/registered/scheduler/test_scheduler_control.py
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     run_and_check_memory_leak,
 )
 
-register_cuda_ci(est_time=360, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=367, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=300, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/sessions/test_session_control.py b/test/registered/sessions/test_session_control.py
index 386d57b9b..ed5855d89 100644
--- a/test/registered/sessions/test_session_control.py
+++ b/test/registered/sessions/test_session_control.py
@@ -24,7 +24,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=77, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=87, suite="stage-b-test-1-gpu-large")
 
 
 def remove_prefix(text: str, prefix: str) -> str:
diff --git a/test/registered/sessions/test_streaming_session.py b/test/registered/sessions/test_streaming_session.py
index fc095c055..f565f7742 100644
--- a/test/registered/sessions/test_streaming_session.py
+++ b/test/registered/sessions/test_streaming_session.py
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=560, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=691, suite="stage-b-test-1-gpu-large")
 
 LOGPROB_PROMPTS = [
     "The quick brown fox jumps over the lazy dog.",
diff --git a/test/registered/sessions/test_streaming_session_swa.py b/test/registered/sessions/test_streaming_session_swa.py
index 465b26a24..76e124783 100644
--- a/test/registered/sessions/test_streaming_session_swa.py
+++ b/test/registered/sessions/test_streaming_session_swa.py
@@ -24,7 +24,7 @@ from test_streaming_session import (  # noqa: E402
     TestStreamingSessionAbortLeakRepro,
 )
 
-register_cuda_ci(est_time=450, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=519, suite="stage-b-test-1-gpu-large")
 
 
 SWA_MODEL = "openai/gpt-oss-20b"
diff --git a/test/registered/spec/dflash/test_dflash.py b/test/registered/spec/dflash/test_dflash.py
index 1583bd92b..d393e4019 100644
--- a/test/registered/spec/dflash/test_dflash.py
+++ b/test/registered/spec/dflash/test_dflash.py
@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=270, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=302, suite="stage-b-test-1-gpu-small")
 
 
 class TestDFlashServerBase(CustomTestCase, MatchedStopMixin, GSM8KMixin):
diff --git a/test/registered/spec/eagle/test_adaptive_speculative.py b/test/registered/spec/eagle/test_adaptive_speculative.py
index 642f941df..6863eacb4 100644
--- a/test/registered/spec/eagle/test_adaptive_speculative.py
+++ b/test/registered/spec/eagle/test_adaptive_speculative.py
@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=420, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=76, suite="stage-b-test-1-gpu-large")
 
 HIGH_ACCEPT_PROMPT = (
     "Output exactly 128 new lines. "
diff --git a/test/registered/spec/eagle/test_deepseek_v3_fp4_mtp_small.py b/test/registered/spec/eagle/test_deepseek_v3_fp4_mtp_small.py
index bea968024..a8372c84e 100644
--- a/test/registered/spec/eagle/test_deepseek_v3_fp4_mtp_small.py
+++ b/test/registered/spec/eagle/test_deepseek_v3_fp4_mtp_small.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=510, suite="stage-b-test-4-gpu-b200")
+register_cuda_ci(est_time=420, suite="stage-b-test-4-gpu-b200")
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
 SERVER_LAUNCH_TIMEOUT = 1200
diff --git a/test/registered/spec/eagle/test_eagle3_basic.py b/test/registered/spec/eagle/test_eagle3_basic.py
index 19486829e..90e672ef1 100644
--- a/test/registered/spec/eagle/test_eagle3_basic.py
+++ b/test/registered/spec/eagle/test_eagle3_basic.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     DEFAULT_TARGET_MODEL_EAGLE3,
 )
 
-register_cuda_ci(est_time=70, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=88, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=50, suite="stage-b-test-1-gpu-small")
 
 _is_hip = is_hip()
diff --git a/test/registered/spec/eagle/test_eagle_constrained_decoding.py b/test/registered/spec/eagle/test_eagle_constrained_decoding.py
index 13712d152..05536be5e 100644
--- a/test/registered/spec/eagle/test_eagle_constrained_decoding.py
+++ b/test/registered/spec/eagle/test_eagle_constrained_decoding.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=96, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=116, suite="stage-b-test-1-gpu-large")
 
 
 class TestEagleConstrainedDecoding(
diff --git a/test/registered/spec/eagle/test_eagle_dp_attention.py b/test/registered/spec/eagle/test_eagle_dp_attention.py
index 3f9dc9d1c..4fa87f1d5 100644
--- a/test/registered/spec/eagle/test_eagle_dp_attention.py
+++ b/test/registered/spec/eagle/test_eagle_dp_attention.py
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
 )
 
 # EAGLE3 with DP attention (tp=2, dp=2, requires 4 GPUs)
-register_cuda_ci(est_time=96, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=99, suite="stage-c-test-4-gpu-h100")
 register_amd_ci(est_time=200, suite="stage-c-test-4-gpu-amd")
 
 
diff --git a/test/registered/spec/eagle/test_eagle_infer_a.py b/test/registered/spec/eagle/test_eagle_infer_a.py
index e7bfd9936..eca84327c 100644
--- a/test/registered/spec/eagle/test_eagle_infer_a.py
+++ b/test/registered/spec/eagle/test_eagle_infer_a.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
-register_cuda_ci(est_time=266, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=357, suite="stage-b-test-1-gpu-large")
 
 
 class TestEAGLEEngine(CustomTestCase):
diff --git a/test/registered/spec/eagle/test_eagle_infer_b.py b/test/registered/spec/eagle/test_eagle_infer_b.py
index c603a04d8..7e941b38e 100644
--- a/test/registered/spec/eagle/test_eagle_infer_b.py
+++ b/test/registered/spec/eagle/test_eagle_infer_b.py
@@ -22,7 +22,7 @@ from sglang.test.run_eval import run_eval
 from sglang.test.server_fixtures.eagle_fixture import EagleServerBase
 from sglang.test.test_utils import DEFAULT_TARGET_MODEL_EAGLE, run_logprob_check
 
-register_cuda_ci(est_time=684, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=847, suite="stage-b-test-1-gpu-large")
 
 
 class TestEAGLEServerBasic(EagleServerBase):
diff --git a/test/registered/spec/eagle/test_eagle_infer_beta.py b/test/registered/spec/eagle/test_eagle_infer_beta.py
index 77a571c14..96a2096a7 100644
--- a/test/registered/spec/eagle/test_eagle_infer_beta.py
+++ b/test/registered/spec/eagle/test_eagle_infer_beta.py
@@ -19,7 +19,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=333, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=369, suite="stage-b-test-1-gpu-small")
 
 
 class TestEagle3ServerBase(CustomTestCase, MatchedStopMixin):
diff --git a/test/registered/spec/eagle/test_eagle_infer_beta_dp_attention.py b/test/registered/spec/eagle/test_eagle_infer_beta_dp_attention.py
index ba21aabc3..bb31b88ae 100644
--- a/test/registered/spec/eagle/test_eagle_infer_beta_dp_attention.py
+++ b/test/registered/spec/eagle/test_eagle_infer_beta_dp_attention.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
 )
 
 # EAGLE with DP attention on B200 (tp=2, dp=2, requires 4 B200 GPUs)
-register_cuda_ci(est_time=136, suite="stage-c-test-4-gpu-b200-small")
+register_cuda_ci(est_time=123, suite="stage-c-test-4-gpu-b200-small")
 
 
 def test_gsm8k(base_url: str, model: str):
diff --git a/test/registered/spec/test_constrained_decoding_spec_reasoning.py b/test/registered/spec/test_constrained_decoding_spec_reasoning.py
index 81e53d01e..18225dc25 100644
--- a/test/registered/spec/test_constrained_decoding_spec_reasoning.py
+++ b/test/registered/spec/test_constrained_decoding_spec_reasoning.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
 )
 
 # Constrained decoding with EAGLE3 speculative reasoning (tp=2)
-register_cuda_ci(est_time=80, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=137, suite="stage-b-test-2-gpu-large")
 
 
 class ServerWithGrammar(CustomTestCase):
diff --git a/test/registered/spec/test_ngram_speculative_decoding.py b/test/registered/spec/test_ngram_speculative_decoding.py
index f1fdfb11c..ad3904b7d 100644
--- a/test/registered/spec/test_ngram_speculative_decoding.py
+++ b/test/registered/spec/test_ngram_speculative_decoding.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=200, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=254, suite="stage-b-test-1-gpu-large")
 
 GSM_DATASET_PATH = None
 
diff --git a/test/registered/spec/test_standalone_speculative_decoding.py b/test/registered/spec/test_standalone_speculative_decoding.py
index cebe13467..6510210df 100644
--- a/test/registered/spec/test_standalone_speculative_decoding.py
+++ b/test/registered/spec/test_standalone_speculative_decoding.py
@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
 )
 
 # Standalone speculative decoding tests (FA3, Triton, FlashInfer backends)
-register_cuda_ci(est_time=294, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=406, suite="stage-b-test-1-gpu-large")
 
 GSM_DATASET_PATH = None
 
diff --git a/test/registered/spec/utils/test_build_eagle_tree.py b/test/registered/spec/utils/test_build_eagle_tree.py
index d07afeb7c..103349fcc 100644
--- a/test/registered/spec/utils/test_build_eagle_tree.py
+++ b/test/registered/spec/utils/test_build_eagle_tree.py
@@ -9,7 +9,7 @@ from sglang.srt.speculative.eagle_utils import (
 from sglang.srt.utils import get_device
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=4, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=6, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=3, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/tokenizer/test_multi_tokenizer.py b/test/registered/tokenizer/test_multi_tokenizer.py
index caa0dd2cb..134bcf845 100644
--- a/test/registered/tokenizer/test_multi_tokenizer.py
+++ b/test/registered/tokenizer/test_multi_tokenizer.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=204, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=211, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=345, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/tokenizer/test_skip_tokenizer_init.py b/test/registered/tokenizer/test_skip_tokenizer_init.py
index 3cb491979..89a5ff977 100644
--- a/test/registered/tokenizer/test_skip_tokenizer_init.py
+++ b/test/registered/tokenizer/test_skip_tokenizer_init.py
@@ -23,7 +23,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=82, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=79, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=117, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/unit/auto_benchmark/test_dataset_tools.py b/test/registered/unit/auto_benchmark/test_dataset_tools.py
index 821921289..cab0aaee3 100644
--- a/test/registered/unit/auto_benchmark/test_dataset_tools.py
+++ b/test/registered/unit/auto_benchmark/test_dataset_tools.py
@@ -15,7 +15,7 @@ from sglang.auto_benchmark_lib import infer_backend, prepare_dataset
 from sglang.benchmark.datasets.autobench import sample_autobench_requests
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=5, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=6, suite="stage-b-test-1-gpu-small")
 
 
 class TestAutoBenchmarkDatasetTools(AutoBenchmarkTestCase):
diff --git a/test/registered/unit/auto_benchmark/test_run_candidate.py b/test/registered/unit/auto_benchmark/test_run_candidate.py
index 7c3b78f22..e172fac3e 100644
--- a/test/registered/unit/auto_benchmark/test_run_candidate.py
+++ b/test/registered/unit/auto_benchmark/test_run_candidate.py
@@ -14,7 +14,7 @@ from auto_benchmark import AutoBenchmarkTestCase
 from sglang.auto_benchmark_lib import SearchDeadlineExceeded, run_candidate
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=5, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=6, suite="stage-b-test-1-gpu-small")
 
 
 class TestAutoBenchmarkRunCandidate(AutoBenchmarkTestCase):
diff --git a/test/registered/unit/auto_benchmark/test_search_tools.py b/test/registered/unit/auto_benchmark/test_search_tools.py
index 409d7e2f7..24af47d0b 100644
--- a/test/registered/unit/auto_benchmark/test_search_tools.py
+++ b/test/registered/unit/auto_benchmark/test_search_tools.py
@@ -28,7 +28,7 @@ from sglang.auto_benchmark_lib import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=5, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=6, suite="stage-b-test-1-gpu-small")
 
 
 class TestAutoBenchmarkSearchTools(AutoBenchmarkTestCase):
diff --git a/test/registered/unit/configs/test_linear_attn_model_registry.py b/test/registered/unit/configs/test_linear_attn_model_registry.py
index 8933a25d5..9eb6970ca 100644
--- a/test/registered/unit/configs/test_linear_attn_model_registry.py
+++ b/test/registered/unit/configs/test_linear_attn_model_registry.py
@@ -13,7 +13,7 @@ from sglang.srt.configs.linear_attn_model_registry import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 # Dummy config classes for testing
diff --git a/test/registered/unit/entrypoints/openai/test_matched_stop.py b/test/registered/unit/entrypoints/openai/test_matched_stop.py
index d3d8c65d1..9c0f5d5ee 100644
--- a/test/registered/unit/entrypoints/openai/test_matched_stop.py
+++ b/test/registered/unit/entrypoints/openai/test_matched_stop.py
@@ -3,7 +3,7 @@ import unittest
 from sglang.srt.sampling.sampling_params import MAX_LEN, get_max_seq_length
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=2, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 
 class TestRegexPatternMaxLength(unittest.TestCase):
diff --git a/test/registered/unit/entrypoints/openai/test_protocol.py b/test/registered/unit/entrypoints/openai/test_protocol.py
index d29a1cf1b..ee978f6a7 100644
--- a/test/registered/unit/entrypoints/openai/test_protocol.py
+++ b/test/registered/unit/entrypoints/openai/test_protocol.py
@@ -32,7 +32,7 @@ from sglang.srt.entrypoints.openai.protocol import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=2, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 class TestModelCard(unittest.TestCase):
diff --git a/test/registered/unit/entrypoints/openai/test_serving_chat.py b/test/registered/unit/entrypoints/openai/test_serving_chat.py
index 1fbc8bad7..19f30cf08 100644
--- a/test/registered/unit/entrypoints/openai/test_serving_chat.py
+++ b/test/registered/unit/entrypoints/openai/test_serving_chat.py
@@ -31,7 +31,7 @@ from sglang.srt.managers.io_struct import GenerateReqInput
 from sglang.srt.utils import get_or_create_event_loop
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=8, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=11, suite="stage-a-test-cpu")
 
 
 class _MockTokenizerManager:
diff --git a/test/registered/unit/entrypoints/openai/test_serving_completions.py b/test/registered/unit/entrypoints/openai/test_serving_completions.py
index 7b19b59ca..f090cb4da 100644
--- a/test/registered/unit/entrypoints/openai/test_serving_completions.py
+++ b/test/registered/unit/entrypoints/openai/test_serving_completions.py
@@ -22,7 +22,7 @@ from sglang.srt.managers.tokenizer_manager import TokenizerManager
 from sglang.srt.utils import get_or_create_event_loop
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=8, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=11, suite="stage-a-test-cpu")
 
 
 class _MockTemplateManager:
diff --git a/test/registered/unit/entrypoints/openai/test_serving_embedding.py b/test/registered/unit/entrypoints/openai/test_serving_embedding.py
index ccbd03435..abbec63ec 100644
--- a/test/registered/unit/entrypoints/openai/test_serving_embedding.py
+++ b/test/registered/unit/entrypoints/openai/test_serving_embedding.py
@@ -54,7 +54,7 @@ from sglang.srt.entrypoints.openai.serving_embedding import OpenAIServingEmbeddi
 from sglang.srt.managers.io_struct import EmbeddingReqInput
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=9, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
 
 
 # Mock TokenizerManager for embedding tests
diff --git a/test/registered/unit/entrypoints/test_ssl_cert_refresher.py b/test/registered/unit/entrypoints/test_ssl_cert_refresher.py
index 85393fd86..119a08348 100644
--- a/test/registered/unit/entrypoints/test_ssl_cert_refresher.py
+++ b/test/registered/unit/entrypoints/test_ssl_cert_refresher.py
@@ -8,7 +8,7 @@ from sglang.srt.entrypoints.ssl_utils import SSLCertRefresher
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=13, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=14, suite="stage-a-test-cpu")
 
 
 def _make_temp_pem(content: bytes) -> str:
diff --git a/test/registered/unit/entrypoints/test_v1_loads_aggregate.py b/test/registered/unit/entrypoints/test_v1_loads_aggregate.py
index 75e1cf50c..4273f9687 100644
--- a/test/registered/unit/entrypoints/test_v1_loads_aggregate.py
+++ b/test/registered/unit/entrypoints/test_v1_loads_aggregate.py
@@ -11,7 +11,7 @@ from sglang.srt.entrypoints.v1_loads import _compute_aggregate
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=1, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
 
 
 def _load(
diff --git a/test/registered/unit/eplb/test_balanced_packing.py b/test/registered/unit/eplb/test_balanced_packing.py
index 447e7cdd2..efa94f94d 100644
--- a/test/registered/unit/eplb/test_balanced_packing.py
+++ b/test/registered/unit/eplb/test_balanced_packing.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=5, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 import unittest
 
diff --git a/test/registered/unit/eplb/test_compute_logical_to_rank_dispatch_physical_map.py b/test/registered/unit/eplb/test_compute_logical_to_rank_dispatch_physical_map.py
index 68e75fab2..f4981060c 100644
--- a/test/registered/unit/eplb/test_compute_logical_to_rank_dispatch_physical_map.py
+++ b/test/registered/unit/eplb/test_compute_logical_to_rank_dispatch_physical_map.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=5, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 import types
 import unittest
diff --git a/test/registered/unit/function_call/test_hunyuan_detector.py b/test/registered/unit/function_call/test_hunyuan_detector.py
index 61bef8f01..5b64aac8b 100644
--- a/test/registered/unit/function_call/test_hunyuan_detector.py
+++ b/test/registered/unit/function_call/test_hunyuan_detector.py
@@ -8,7 +8,7 @@ from sglang.srt.function_call.hunyuan_detector import HunyuanDetector
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=1, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 def _make_tools():
diff --git a/test/registered/unit/layers/test_conv_layer.py b/test/registered/unit/layers/test_conv_layer.py
index 20ab6698a..2dedbc3b5 100644
--- a/test/registered/unit/layers/test_conv_layer.py
+++ b/test/registered/unit/layers/test_conv_layer.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=4, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=7, suite="stage-b-test-1-gpu-small")
 
 import unittest
 
diff --git a/test/registered/unit/layers/test_mamba_state_scatter_triton.py b/test/registered/unit/layers/test_mamba_state_scatter_triton.py
index c1f106c4b..1f20373b3 100644
--- a/test/registered/unit/layers/test_mamba_state_scatter_triton.py
+++ b/test/registered/unit/layers/test_mamba_state_scatter_triton.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=4, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=7, suite="stage-b-test-1-gpu-small")
 
 import os
 import unittest
diff --git a/test/registered/unit/layers/test_pooler_score_and_pool.py b/test/registered/unit/layers/test_pooler_score_and_pool.py
index 829b228ba..9ce8d33ff 100644
--- a/test/registered/unit/layers/test_pooler_score_and_pool.py
+++ b/test/registered/unit/layers/test_pooler_score_and_pool.py
@@ -19,7 +19,7 @@ from sglang.srt.layers.pooler import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=9, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
 
 
 def _make_forward_batch(
diff --git a/test/registered/unit/lora/test_mem_pool_ep_unit.py b/test/registered/unit/lora/test_mem_pool_ep_unit.py
index b37248c30..6a374e058 100644
--- a/test/registered/unit/lora/test_mem_pool_ep_unit.py
+++ b/test/registered/unit/lora/test_mem_pool_ep_unit.py
@@ -16,7 +16,7 @@ Usage:
 from sglang.test.ci.ci_register import register_cuda_ci
 
 # CPU-only unit test; no CUDA/distributed dependencies.
-register_cuda_ci(est_time=4, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
 
 import types
 import unittest
diff --git a/test/registered/unit/managers/test_dp_budget.py b/test/registered/unit/managers/test_dp_budget.py
index ccf2f8449..5d4e66cea 100644
--- a/test/registered/unit/managers/test_dp_budget.py
+++ b/test/registered/unit/managers/test_dp_budget.py
@@ -17,7 +17,7 @@ maybe_stub_sgl_kernel()
 from sglang.srt.managers.data_parallel_controller import DPBudget
 from sglang.srt.managers.io_struct import GetLoadsReqOutput, WatchLoadUpdateReq
 
-register_cpu_ci(est_time=2, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=11, suite="stage-a-test-cpu")
 
 
 _BASE_LOAD = GetLoadsReqOutput(
diff --git a/test/registered/unit/managers/test_hisparse_unit.py b/test/registered/unit/managers/test_hisparse_unit.py
index 6c4028be6..e3f3d54a5 100644
--- a/test/registered/unit/managers/test_hisparse_unit.py
+++ b/test/registered/unit/managers/test_hisparse_unit.py
@@ -16,7 +16,7 @@ import torch
 from sglang.srt.utils import is_cuda, is_hip, is_npu, is_xpu
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=20, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=10, suite="stage-b-test-1-gpu-small")
 
 # ---------------------------------------------------------------------------
 # Test configuration (small-scale for fast CI runs)
diff --git a/test/registered/unit/managers/test_prefill_adder.py b/test/registered/unit/managers/test_prefill_adder.py
index 85a4acf39..182b78938 100644
--- a/test/registered/unit/managers/test_prefill_adder.py
+++ b/test/registered/unit/managers/test_prefill_adder.py
@@ -12,7 +12,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=2, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/unit/managers/test_scheduler_flush_cache.py b/test/registered/unit/managers/test_scheduler_flush_cache.py
index 249b4b8ce..0b5d5aadd 100644
--- a/test/registered/unit/managers/test_scheduler_flush_cache.py
+++ b/test/registered/unit/managers/test_scheduler_flush_cache.py
@@ -9,7 +9,7 @@ maybe_stub_sgl_kernel()
 from sglang.srt.managers.io_struct import FlushCacheReqInput
 from sglang.srt.managers.scheduler import Scheduler
 
-register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=14, suite="stage-a-test-cpu")
 
 
 class TestSchedulerFlushCache(unittest.TestCase):
diff --git a/test/registered/unit/managers/test_scheduler_pause_generation.py b/test/registered/unit/managers/test_scheduler_pause_generation.py
index 46bc664eb..b1059a33e 100644
--- a/test/registered/unit/managers/test_scheduler_pause_generation.py
+++ b/test/registered/unit/managers/test_scheduler_pause_generation.py
@@ -11,7 +11,7 @@ from sglang.srt.managers.io_struct import PauseGenerationReqInput
 from sglang.srt.managers.scheduler import Scheduler
 from sglang.srt.managers.scheduler_runtime_checker_mixin import PoolStats
 
-register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=15, suite="stage-a-test-cpu")
 
 
 class TestSchedulerPauseGeneration(unittest.TestCase):
diff --git a/test/registered/unit/mem_cache/test_evict_policy.py b/test/registered/unit/mem_cache/test_evict_policy.py
index 13e022acf..8365c0580 100644
--- a/test/registered/unit/mem_cache/test_evict_policy.py
+++ b/test/registered/unit/mem_cache/test_evict_policy.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=1, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 import unittest
 from unittest.mock import MagicMock
diff --git a/test/registered/unit/mem_cache/test_mamba_unittest.py b/test/registered/unit/mem_cache/test_mamba_unittest.py
index bc7c75a9a..6158f622b 100755
--- a/test/registered/unit/mem_cache/test_mamba_unittest.py
+++ b/test/registered/unit/mem_cache/test_mamba_unittest.py
@@ -21,7 +21,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.srt.utils import get_device
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=10, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=9, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/unit/mem_cache/test_nsa_pool_host_unit.py b/test/registered/unit/mem_cache/test_nsa_pool_host_unit.py
index 2602e196a..6cac74027 100644
--- a/test/registered/unit/mem_cache/test_nsa_pool_host_unit.py
+++ b/test/registered/unit/mem_cache/test_nsa_pool_host_unit.py
@@ -12,7 +12,7 @@ from sglang.srt.mem_cache.memory_pool_host import (
 from sglang.srt.utils import is_cuda, is_hip, is_npu, is_xpu
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
 
 
 class TestNSAHiCacheTransfer(unittest.TestCase):
diff --git a/test/registered/unit/mem_cache/test_radix_cache_slru_accuracy.py b/test/registered/unit/mem_cache/test_radix_cache_slru_accuracy.py
index bc28b40b7..46dc341a3 100644
--- a/test/registered/unit/mem_cache/test_radix_cache_slru_accuracy.py
+++ b/test/registered/unit/mem_cache/test_radix_cache_slru_accuracy.py
@@ -13,7 +13,7 @@ from sglang.srt.mem_cache.memory_pool import MHATokenToKVPool, ReqToTokenPool
 from sglang.srt.mem_cache.radix_cache import RadixCache, RadixKey
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=7, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
 
 
 class TestSLRUAccuracy(unittest.TestCase):
diff --git a/test/registered/unit/mem_cache/test_radix_cache_unit.py b/test/registered/unit/mem_cache/test_radix_cache_unit.py
index c45e3cef8..73ff4fd13 100644
--- a/test/registered/unit/mem_cache/test_radix_cache_unit.py
+++ b/test/registered/unit/mem_cache/test_radix_cache_unit.py
@@ -21,7 +21,7 @@ from sglang.srt.mem_cache.common import available_and_evictable_str
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 # CPU-based unit test, runs quickly on any GPU runner
-register_cuda_ci(est_time=16, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=15, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=5, suite="stage-b-test-1-gpu-small-amd")
 
 import random
diff --git a/test/registered/unit/mem_cache/test_streaming_session_unit.py b/test/registered/unit/mem_cache/test_streaming_session_unit.py
index b4ac9e8c3..7960470bb 100644
--- a/test/registered/unit/mem_cache/test_streaming_session_unit.py
+++ b/test/registered/unit/mem_cache/test_streaming_session_unit.py
@@ -7,7 +7,7 @@ from sglang.srt.mem_cache.base_prefix_cache import MatchResult
 from sglang.srt.session.streaming_session import SessionSlot, StreamingSession
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=8, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=12, suite="stage-a-test-cpu")
 
 
 class _FakeAllocator:
diff --git a/test/registered/unit/mem_cache/test_swa_eviction_boundary.py b/test/registered/unit/mem_cache/test_swa_eviction_boundary.py
index b4c302261..751dccee2 100644
--- a/test/registered/unit/mem_cache/test_swa_eviction_boundary.py
+++ b/test/registered/unit/mem_cache/test_swa_eviction_boundary.py
@@ -26,7 +26,7 @@ from sglang.srt.mem_cache.swa_radix_cache import SWARadixCache
 from sglang.srt.utils import get_device
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=12, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=10, suite="stage-b-test-1-gpu-small-amd")
 
 # ---------------------------------------------------------------------------
diff --git a/test/registered/unit/mem_cache/test_swa_unittest.py b/test/registered/unit/mem_cache/test_swa_unittest.py
index e5b922166..d3a6e2c42 100644
--- a/test/registered/unit/mem_cache/test_swa_unittest.py
+++ b/test/registered/unit/mem_cache/test_swa_unittest.py
@@ -17,7 +17,7 @@ from sglang.srt.mem_cache.swa_radix_cache import SWARadixCache
 from sglang.srt.utils import get_device
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=10, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/unit/mem_cache/test_unified_radix_cache_bench.py b/test/registered/unit/mem_cache/test_unified_radix_cache_bench.py
index 5dce0295a..f775b213a 100644
--- a/test/registered/unit/mem_cache/test_unified_radix_cache_bench.py
+++ b/test/registered/unit/mem_cache/test_unified_radix_cache_bench.py
@@ -38,7 +38,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.srt.utils import get_device
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=25, suite="stage-b-test-1-gpu-small")
 
 # ---------------------------------------------------------------------------
 # Constants
diff --git a/test/registered/unit/mem_cache/test_unified_radix_cache_unittest.py b/test/registered/unit/mem_cache/test_unified_radix_cache_unittest.py
index 72d9da236..114a06e1b 100644
--- a/test/registered/unit/mem_cache/test_unified_radix_cache_unittest.py
+++ b/test/registered/unit/mem_cache/test_unified_radix_cache_unittest.py
@@ -39,7 +39,7 @@ from sglang.srt.utils import get_device
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=30, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=10, suite="stage-b-test-1-gpu-small")
 
 
 @dataclass(frozen=True)
diff --git a/test/registered/unit/model_executor/test_pool_configurator.py b/test/registered/unit/model_executor/test_pool_configurator.py
index 04dc59e12..ea98b2c4d 100644
--- a/test/registered/unit/model_executor/test_pool_configurator.py
+++ b/test/registered/unit/model_executor/test_pool_configurator.py
@@ -12,7 +12,7 @@ from unittest.mock import MagicMock, patch
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=9, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
 
 
 @contextlib.contextmanager
diff --git a/test/registered/unit/model_loader/test_modelopt_export.py b/test/registered/unit/model_loader/test_modelopt_export.py
index 697af4a2c..a1eb133c2 100644
--- a/test/registered/unit/model_loader/test_modelopt_export.py
+++ b/test/registered/unit/model_loader/test_modelopt_export.py
@@ -19,7 +19,7 @@ from sglang.srt.configs.model_config import ModelConfig
 from sglang.srt.model_loader.loader import ModelOptModelLoader
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=9, suite="stage-b-test-1-gpu-small-amd")
 
 # Note: PYTHONPATH=python should be set when running tests
diff --git a/test/registered/unit/model_loader/test_modelopt_loader.py b/test/registered/unit/model_loader/test_modelopt_loader.py
index a7afcaf24..7f9652c0e 100644
--- a/test/registered/unit/model_loader/test_modelopt_loader.py
+++ b/test/registered/unit/model_loader/test_modelopt_loader.py
@@ -30,7 +30,7 @@ CALIBRATION_BATCH_SIZE = 36
 CALIBRATION_NUM_SAMPLES = 512
 DEFAULT_DEVICE = "cuda:0"
 
-register_cuda_ci(est_time=10, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-small")
 
 
 class TestModelOptModelLoader(CustomTestCase):
diff --git a/test/registered/unit/model_loader/test_prefetch_checkpoints.py b/test/registered/unit/model_loader/test_prefetch_checkpoints.py
index 9d234df2c..0a3e91faf 100644
--- a/test/registered/unit/model_loader/test_prefetch_checkpoints.py
+++ b/test/registered/unit/model_loader/test_prefetch_checkpoints.py
@@ -18,7 +18,7 @@ from sglang.srt.model_loader.weight_utils import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=5, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
 
 
 class TestPrefetchWeightsIdentical(unittest.TestCase):
diff --git a/test/registered/unit/models/test_llava.py b/test/registered/unit/models/test_llava.py
index 1e4ca1ce2..4ebaecf8e 100644
--- a/test/registered/unit/models/test_llava.py
+++ b/test/registered/unit/models/test_llava.py
@@ -5,7 +5,7 @@ from sglang.srt.models.llava import AutoModel, LlavaForConditionalGeneration
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
 
 
 class PixtralVisionConfig:
diff --git a/test/registered/unit/observability/test_func_timer.py b/test/registered/unit/observability/test_func_timer.py
index 988a5ca27..30608ca8e 100644
--- a/test/registered/unit/observability/test_func_timer.py
+++ b/test/registered/unit/observability/test_func_timer.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=1, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 import asyncio
 import unittest
diff --git a/test/registered/unit/observability/test_label_transform.py b/test/registered/unit/observability/test_label_transform.py
index 2b941006a..f71c66bd1 100644
--- a/test/registered/unit/observability/test_label_transform.py
+++ b/test/registered/unit/observability/test_label_transform.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=1, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 import unittest
 
diff --git a/test/registered/unit/observability/test_metrics_utils.py b/test/registered/unit/observability/test_metrics_utils.py
index 1d20fd944..6eb26d8ff 100644
--- a/test/registered/unit/observability/test_metrics_utils.py
+++ b/test/registered/unit/observability/test_metrics_utils.py
@@ -6,7 +6,7 @@ from sglang.srt.observability.utils import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=1, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 
 class TestMetricsUtils(unittest.TestCase):
diff --git a/test/registered/unit/observability/test_request_metrics_exporter.py b/test/registered/unit/observability/test_request_metrics_exporter.py
index 3dd4f618c..683bb10b6 100644
--- a/test/registered/unit/observability/test_request_metrics_exporter.py
+++ b/test/registered/unit/observability/test_request_metrics_exporter.py
@@ -5,7 +5,7 @@ from typing import Any, Dict, List, Optional
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=1, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 import asyncio
 import json
diff --git a/test/registered/unit/observability/test_startup_func_log_and_timer.py b/test/registered/unit/observability/test_startup_func_log_and_timer.py
index 5afcfab8f..90fcf5074 100644
--- a/test/registered/unit/observability/test_startup_func_log_and_timer.py
+++ b/test/registered/unit/observability/test_startup_func_log_and_timer.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=1, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 import unittest
 from unittest.mock import MagicMock, patch
diff --git a/test/registered/unit/observability/test_trace.py b/test/registered/unit/observability/test_trace.py
index 78654030c..719f31270 100644
--- a/test/registered/unit/observability/test_trace.py
+++ b/test/registered/unit/observability/test_trace.py
@@ -4,7 +4,7 @@ import os
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=3, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 import threading
 import unittest
diff --git a/test/registered/unit/parser/test_code_completion_parser.py b/test/registered/unit/parser/test_code_completion_parser.py
index 1182e0910..b8a411585 100644
--- a/test/registered/unit/parser/test_code_completion_parser.py
+++ b/test/registered/unit/parser/test_code_completion_parser.py
@@ -18,7 +18,7 @@ from sglang.srt.parser.code_completion_parser import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 class TestFimPosition(CustomTestCase):
diff --git a/test/registered/unit/parser/test_conversation.py b/test/registered/unit/parser/test_conversation.py
index ed0a43c44..72bfb8cca 100644
--- a/test/registered/unit/parser/test_conversation.py
+++ b/test/registered/unit/parser/test_conversation.py
@@ -32,7 +32,7 @@ from sglang.srt.parser.conversation import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 class TestConversationGetPrompt(CustomTestCase):
diff --git a/test/registered/unit/parser/test_harmony_parser.py b/test/registered/unit/parser/test_harmony_parser.py
index 02c559146..26242a9db 100644
--- a/test/registered/unit/parser/test_harmony_parser.py
+++ b/test/registered/unit/parser/test_harmony_parser.py
@@ -14,7 +14,7 @@ from sglang.srt.parser.harmony_parser import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 class TestEvent(CustomTestCase):
diff --git a/test/registered/unit/parser/test_jinja_template_utils.py b/test/registered/unit/parser/test_jinja_template_utils.py
index c3845148e..3396de98a 100644
--- a/test/registered/unit/parser/test_jinja_template_utils.py
+++ b/test/registered/unit/parser/test_jinja_template_utils.py
@@ -9,7 +9,7 @@ from sglang.srt.parser.jinja_template_utils import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 class TestTemplateContentFormatDetection(CustomTestCase):
diff --git a/test/registered/unit/parser/test_reasoning_content_without_parser.py b/test/registered/unit/parser/test_reasoning_content_without_parser.py
index 265adb711..f8eada591 100644
--- a/test/registered/unit/parser/test_reasoning_content_without_parser.py
+++ b/test/registered/unit/parser/test_reasoning_content_without_parser.py
@@ -4,7 +4,7 @@ from sglang.srt.parser.reasoning_parser import ReasoningParser
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 # Simulated model output that contains think tags (e.g. from DeepSeek-R1)
 THINK_OUTPUT = (
diff --git a/test/registered/unit/parser/test_reasoning_parser.py b/test/registered/unit/parser/test_reasoning_parser.py
index f33c88414..958168a7d 100644
--- a/test/registered/unit/parser/test_reasoning_parser.py
+++ b/test/registered/unit/parser/test_reasoning_parser.py
@@ -18,7 +18,7 @@ from sglang.srt.parser.reasoning_parser import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 class TestStreamingParseResult(CustomTestCase):
diff --git a/test/registered/unit/platforms/test_platform_interface.py b/test/registered/unit/platforms/test_platform_interface.py
index e77a83c6f..f5bfcd337 100644
--- a/test/registered/unit/platforms/test_platform_interface.py
+++ b/test/registered/unit/platforms/test_platform_interface.py
@@ -18,7 +18,7 @@ from sglang.srt.platforms.interface import SRTPlatform
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 # ---------------------------------------------------------------------------
diff --git a/test/registered/unit/plugins/test_hook_registry.py b/test/registered/unit/plugins/test_hook_registry.py
index 4f066189a..b01ed3b37 100644
--- a/test/registered/unit/plugins/test_hook_registry.py
+++ b/test/registered/unit/plugins/test_hook_registry.py
@@ -16,7 +16,7 @@ from sglang.srt.plugins.hook_registry import HookRegistry, HookType, plugin_hook
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 # ---------------------------------------------------------------------------
 # Helpers: synthetic module creation
diff --git a/test/registered/unit/plugins/test_load_plugins.py b/test/registered/unit/plugins/test_load_plugins.py
index 218f0fc27..be2e13570 100644
--- a/test/registered/unit/plugins/test_load_plugins.py
+++ b/test/registered/unit/plugins/test_load_plugins.py
@@ -19,7 +19,7 @@ from sglang.srt.plugins import (
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 def _make_ep(name, dist_name=None, load_fn=None):
diff --git a/test/registered/unit/sampling/test_custom_logit_processor.py b/test/registered/unit/sampling/test_custom_logit_processor.py
index 6ef5544aa..eb9a76d0c 100644
--- a/test/registered/unit/sampling/test_custom_logit_processor.py
+++ b/test/registered/unit/sampling/test_custom_logit_processor.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 import json
 import unittest
diff --git a/test/registered/unit/sampling/test_penaltylib.py b/test/registered/unit/sampling/test_penaltylib.py
index 4943ff887..5e39ed32c 100644
--- a/test/registered/unit/sampling/test_penaltylib.py
+++ b/test/registered/unit/sampling/test_penaltylib.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=8, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=9, suite="stage-a-test-cpu")
 
 import unittest
 from unittest.mock import MagicMock
diff --git a/test/registered/unit/sampling/test_sampling_batch_info.py b/test/registered/unit/sampling/test_sampling_batch_info.py
index 9af7587f2..c3df29892 100644
--- a/test/registered/unit/sampling/test_sampling_batch_info.py
+++ b/test/registered/unit/sampling/test_sampling_batch_info.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=8, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=9, suite="stage-a-test-cpu")
 
 import unittest
 from unittest.mock import MagicMock, patch
diff --git a/test/registered/unit/sampling/test_sampling_params.py b/test/registered/unit/sampling/test_sampling_params.py
index ce9a7c3c3..b849e6e85 100644
--- a/test/registered/unit/sampling/test_sampling_params.py
+++ b/test/registered/unit/sampling/test_sampling_params.py
@@ -2,7 +2,7 @@
 
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 import unittest
 from unittest.mock import MagicMock
diff --git a/test/registered/unit/server_args/test_server_args.py b/test/registered/unit/server_args/test_server_args.py
index 29b962b74..a83a2c693 100644
--- a/test/registered/unit/server_args/test_server_args.py
+++ b/test/registered/unit/server_args/test_server_args.py
@@ -10,7 +10,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
 )
 
-register_cpu_ci(est_time=9, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=10, suite="stage-a-test-cpu")
 
 # Mock get_device() so all tests run on CPU-only CI runners
 _mock_device = patch("sglang.srt.server_args.get_device", return_value="cuda")
diff --git a/test/registered/unit/spec/test_adaptive_spec_params.py b/test/registered/unit/spec/test_adaptive_spec_params.py
index 737908249..217064801 100644
--- a/test/registered/unit/spec/test_adaptive_spec_params.py
+++ b/test/registered/unit/spec/test_adaptive_spec_params.py
@@ -3,7 +3,7 @@ import unittest
 from sglang.srt.speculative.adaptive_spec_params import AdaptiveSpeculativeParams
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=2, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 
 class TestAdaptiveSpeculativeParams(unittest.TestCase):
diff --git a/test/registered/unit/spec/test_ngram_corpus.py b/test/registered/unit/spec/test_ngram_corpus.py
index 97b0906e1..8d491b5b7 100644
--- a/test/registered/unit/spec/test_ngram_corpus.py
+++ b/test/registered/unit/spec/test_ngram_corpus.py
@@ -13,7 +13,7 @@ from sglang.srt.speculative.cpp_ngram.ngram_corpus import NgramCorpus
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=18, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=26, suite="stage-a-test-cpu")
 
 
 def _make_corpus(match_type="BFS", **kwargs):
diff --git a/test/registered/unit/test_runai_utils.py b/test/registered/unit/test_runai_utils.py
index 865b3c7a9..83a0adf1a 100644
--- a/test/registered/unit/test_runai_utils.py
+++ b/test/registered/unit/test_runai_utils.py
@@ -6,7 +6,7 @@ from sglang.srt.utils.runai_utils import ObjectStorageModel, is_runai_obj_uri
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 class TestRunaiUtils(CustomTestCase):
diff --git a/test/registered/unit/tokenizer/test_tiktoken_tokenizer.py b/test/registered/unit/tokenizer/test_tiktoken_tokenizer.py
index c683d0178..8b3023317 100644
--- a/test/registered/unit/tokenizer/test_tiktoken_tokenizer.py
+++ b/test/registered/unit/tokenizer/test_tiktoken_tokenizer.py
@@ -6,7 +6,7 @@ from unittest.mock import MagicMock, patch
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 from sglang.srt.tokenizer.tiktoken_tokenizer import (
     CONTROL_TOKEN_TEXTS,
diff --git a/test/registered/unit/tools/test_get_version_tag.py b/test/registered/unit/tools/test_get_version_tag.py
index d64dac347..b6114fef1 100644
--- a/test/registered/unit/tools/test_get_version_tag.py
+++ b/test/registered/unit/tools/test_get_version_tag.py
@@ -33,7 +33,7 @@ def _load_module(name, path):
 
 
 register_cpu_ci = _load_module("ci_register", CI_REGISTER_PATH).register_cpu_ci
-register_cpu_ci(est_time=2, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=0, suite="stage-a-test-cpu")
 
 
 class TestGetVersionTag(unittest.TestCase):
diff --git a/test/registered/unit/utils/test_gauge_histogram.py b/test/registered/unit/utils/test_gauge_histogram.py
index e208d3e7a..b2dac54f9 100644
--- a/test/registered/unit/utils/test_gauge_histogram.py
+++ b/test/registered/unit/utils/test_gauge_histogram.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=3, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 import unittest
 
diff --git a/test/registered/unit/utils/test_hf_transformers.py b/test/registered/unit/utils/test_hf_transformers.py
index cc6226a73..df7fddb29 100644
--- a/test/registered/unit/utils/test_hf_transformers.py
+++ b/test/registered/unit/utils/test_hf_transformers.py
@@ -24,7 +24,7 @@ from sglang.srt.utils.hf_transformers.tokenizer import _fix_special_tokens_patte
 from sglang.srt.utils.hf_transformers_patches import normalize_rope_scaling_compat
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=5, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 
 # ---------------------------------------------------------------------------
diff --git a/test/registered/unit/utils/test_http_server_auth.py b/test/registered/unit/utils/test_http_server_auth.py
index 25c948704..69f50154e 100644
--- a/test/registered/unit/utils/test_http_server_auth.py
+++ b/test/registered/unit/utils/test_http_server_auth.py
@@ -10,7 +10,7 @@ import unittest
 from sglang.srt.utils.auth import AuthLevel, decide_request_auth
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=3, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 
 class TestHttpServerAdminAuth(unittest.TestCase):
diff --git a/test/registered/unit/utils/test_json_response.py b/test/registered/unit/utils/test_json_response.py
index 20669d4ac..764de5f53 100644
--- a/test/registered/unit/utils/test_json_response.py
+++ b/test/registered/unit/utils/test_json_response.py
@@ -10,7 +10,7 @@ from sglang.srt.utils.json_response import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=4, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 
 class TestJSONResponseUtils(unittest.TestCase):
diff --git a/test/registered/unit/utils/test_profile_merger.py b/test/registered/unit/utils/test_profile_merger.py
index e410af33e..29455288d 100644
--- a/test/registered/unit/utils/test_profile_merger.py
+++ b/test/registered/unit/utils/test_profile_merger.py
@@ -17,7 +17,7 @@ from sglang.srt.managers.io_struct import ProfileReq, ProfileReqInput, ProfileRe
 from sglang.srt.utils.profile_merger import ProfileMerger
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=8, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/unit/utils/test_subprocess_watchdog.py b/test/registered/unit/utils/test_subprocess_watchdog.py
index 1e9f50f92..48873ecbe 100644
--- a/test/registered/unit/utils/test_subprocess_watchdog.py
+++ b/test/registered/unit/utils/test_subprocess_watchdog.py
@@ -24,7 +24,7 @@ from sglang.srt.utils.watchdog import SubprocessWatchdog
 from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cpu_ci(est_time=8, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=9, suite="stage-a-test-cpu")
 
 
 def healthy_worker():
diff --git a/test/registered/utils/test_log_utils.py b/test/registered/utils/test_log_utils.py
index 449b8a856..b3007e220 100644
--- a/test/registered/utils/test_log_utils.py
+++ b/test/registered/utils/test_log_utils.py
@@ -9,7 +9,7 @@ from pathlib import Path
 from sglang.srt.utils.log_utils import create_log_targets, log_json
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=3, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
 
 
 class TestLogUtils(unittest.TestCase):
diff --git a/test/registered/utils/test_network_address.py b/test/registered/utils/test_network_address.py
index 529162f2c..ff05206c2 100644
--- a/test/registered/utils/test_network_address.py
+++ b/test/registered/utils/test_network_address.py
@@ -6,7 +6,7 @@ from sglang.srt.server_args import PortArgs, ServerArgs
 from sglang.srt.utils.network import NetworkAddress
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 # Mock get_device() so ServerArgs tests run on CPU-only CI runners
 _mock_device = patch("sglang.srt.server_args.get_device", return_value="cuda")
diff --git a/test/registered/utils/test_numa_utils.py b/test/registered/utils/test_numa_utils.py
index fc3fe61bc..27719d306 100644
--- a/test/registered/utils/test_numa_utils.py
+++ b/test/registered/utils/test_numa_utils.py
@@ -9,7 +9,7 @@ from sglang.srt.utils.numa_utils import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci, register_cuda_ci
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 register_cuda_ci(est_time=10, suite="stage-c-test-4-gpu-gb200")
 register_cuda_ci(est_time=10, suite="stage-c-test-8-gpu-b200")
 
diff --git a/test/registered/utils/test_socket_utils.py b/test/registered/utils/test_socket_utils.py
index 058b81711..63a42ac39 100644
--- a/test/registered/utils/test_socket_utils.py
+++ b/test/registered/utils/test_socket_utils.py
@@ -15,7 +15,7 @@ from sglang.test.ci.ci_register import register_cpu_ci
 from sglang.test.test_utils import CustomTestCase
 from sglang.utils import normalize_base_url, release_port, reserve_port
 
-register_cpu_ci(est_time=6, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 class TestTryBindSocket(CustomTestCase):
diff --git a/test/registered/vlm/test_evs.py b/test/registered/vlm/test_evs.py
index cd4aa8b48..881089abc 100644
--- a/test/registered/vlm/test_evs.py
+++ b/test/registered/vlm/test_evs.py
@@ -6,7 +6,7 @@ import pytest
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import run_doctests
 
-register_cuda_ci(est_time=8, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-small-amd")
 
 
diff --git a/test/registered/vlm/test_video_utils.py b/test/registered/vlm/test_video_utils.py
index f4dd0b027..7587bb9ee 100644
--- a/test/registered/vlm/test_video_utils.py
+++ b/test/registered/vlm/test_video_utils.py
@@ -5,7 +5,7 @@ import pytest
 from sglang.srt.utils import sample_video_frames
 from sglang.test.ci.ci_register import register_cpu_ci
 
-register_cpu_ci(est_time=4, suite="stage-a-test-cpu")
+register_cpu_ci(est_time=7, suite="stage-a-test-cpu")
 
 
 class DummyVideo:
diff --git a/test/registered/vlm/test_vision_chunked_prefill.py b/test/registered/vlm/test_vision_chunked_prefill.py
index 57739821d..762923353 100644
--- a/test/registered/vlm/test_vision_chunked_prefill.py
+++ b/test/registered/vlm/test_vision_chunked_prefill.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=143, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=156, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=270, suite="stage-b-test-1-gpu-small-amd")
 """
 Usage:
diff --git a/test/registered/vlm/test_vision_openai_server_a.py b/test/registered/vlm/test_vision_openai_server_a.py
index c41eb1f61..d7acd2c16 100644
--- a/test/registered/vlm/test_vision_openai_server_a.py
+++ b/test/registered/vlm/test_vision_openai_server_a.py
@@ -19,7 +19,7 @@ from sglang.test.vlm_utils import (
     VideoOpenAITestMixin,
 )
 
-register_cuda_ci(est_time=736, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=780, suite="stage-b-test-1-gpu-large")
 
 
 class TestLlavaServer(ImageOpenAITestMixin):
diff --git a/test/registered/vlm/test_vlm_input_format.py b/test/registered/vlm/test_vlm_input_format.py
index 81dac3550..1a829ae91 100644
--- a/test/registered/vlm/test_vlm_input_format.py
+++ b/test/registered/vlm/test_vlm_input_format.py
@@ -37,7 +37,7 @@ from sglang.srt.entrypoints.openai.protocol import ChatCompletionRequest
 from sglang.srt.parser.conversation import generate_chat_conv
 from sglang.srt.utils.hf_transformers_utils import _fix_added_tokens_encoding
 
-register_cuda_ci(est_time=602, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=747, suite="stage-b-test-1-gpu-large")
 
 IMAGE_MAN_IRONING_URL = "https://raw.githubusercontent.com/sgl-project/sgl-test-files/refs/heads/main/images/man_ironing_on_back_of_suv.png"
 IMAGE_SGL_LOGO_URL = "https://raw.githubusercontent.com/sgl-project/sgl-test-files/refs/heads/main/images/sgl_logo.png"
diff --git a/test/registered/vlm/test_vlm_tp4.py b/test/registered/vlm/test_vlm_tp4.py
index 33137dc00..2ad16ff80 100644
--- a/test/registered/vlm/test_vlm_tp4.py
+++ b/test/registered/vlm/test_vlm_tp4.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=118, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=133, suite="stage-c-test-4-gpu-h100")
 
 QWEN35_27B_MODEL = "Qwen/Qwen3.5-27B"
 MMMU_ACCURACY_THRESHOLD = 0.65