diff --git a/test/registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py b/test/registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py
index 9855a70a8..f658b1b86 100644
--- a/test/registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py
+++ b/test/registered/4-gpu-models/test_nvidia_nemotron_3_super_nvfp4.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=600, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=290, suite="stage-c-test-4-gpu-b200")
 
 NEMOTRON_3_SUPER_NVFP4_MODEL = "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4"
 
diff --git a/test/registered/4-gpu-models/test_qwen35_hicache.py b/test/registered/4-gpu-models/test_qwen35_hicache.py
index fedf461d5..ac001ff32 100644
--- a/test/registered/4-gpu-models/test_qwen35_hicache.py
+++ b/test/registered/4-gpu-models/test_qwen35_hicache.py
@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=600, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=620, suite="stage-c-test-4-gpu-h100")
 
 QWEN35_27B_MODEL = "Qwen/Qwen3.5-27B"
 ACC_THRESHOLDS = {QWEN35_27B_MODEL: {"gsm8k": 0.8}}
diff --git a/test/registered/4-gpu-models/test_qwen3_30b.py b/test/registered/4-gpu-models/test_qwen3_30b.py
index 073723dda..1e7ebbe33 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=300, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=150, 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 b18c463b3..48b711dd7 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=350, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=120, suite="stage-c-test-4-gpu-h100")
 
 QWEN3_NEXT_MODEL = "Qwen/Qwen3-Next-80B-A3B-Instruct"
 
diff --git a/test/registered/4-gpu-models/test_qwen3_next_models_mtp.py b/test/registered/4-gpu-models/test_qwen3_next_models_mtp.py
index 4d04780e2..3be1ddd0c 100644
--- a/test/registered/4-gpu-models/test_qwen3_next_models_mtp.py
+++ b/test/registered/4-gpu-models/test_qwen3_next_models_mtp.py
@@ -7,7 +7,7 @@ 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=500, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=430, 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_v3_mtp.py b/test/registered/8-gpu-models/test_deepseek_v3_mtp.py
index 48e0847ac..e90cd4fb0 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=275, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=240, 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 b2687df9f..3a7b405ea 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=720, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=870, 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 11b311ba8..55457f9fd 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=720, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=880, 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 a32fd3ac5..aa9da353e 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=200, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=270, suite="stage-c-test-8-gpu-h200")
 
 
 class TestMiMoV2Flash(GSM8KMixin, SpecDecodingMixin, DefaultServerBase):
diff --git a/test/registered/8-gpu-models/test_ring_2_5_1t.py b/test/registered/8-gpu-models/test_ring_2_5_1t.py
index 29c64160c..f42fa29f3 100644
--- a/test/registered/8-gpu-models/test_ring_2_5_1t.py
+++ b/test/registered/8-gpu-models/test_ring_2_5_1t.py
@@ -5,7 +5,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.run_combined_tests import run_combined_tests
 from sglang.test.test_utils import ModelLaunchSettings
 
-register_cuda_ci(est_time=1800, suite="nightly-8-gpu-common", nightly=True)
+register_cuda_ci(est_time=510, suite="nightly-8-gpu-common", nightly=True)
 
 RING_2_5_1T_MODEL_PATH = "inclusionAI/Ring-2.5-1T"
 
diff --git a/test/registered/attention/test_fa3.py b/test/registered/attention/test_fa3.py
index 472a2099a..d0c48cfcd 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=300, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=390, suite="stage-b-test-1-gpu-large")
 
 GSM_DATASET_PATH = None
 
diff --git a/test/registered/core/test_deterministic.py b/test/registered/core/test_deterministic.py
index 6d205a369..956504929 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=278, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=200, 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_gpt_oss_1gpu.py b/test/registered/core/test_gpt_oss_1gpu.py
index 179c5e8f3..c208e8077 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=519, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=420, 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 8fc74a9d7..dcbe427e7 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=500, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=330, suite="stage-b-test-1-gpu-small")
 
 
 @unittest.skipIf(not torch.cuda.is_available(), "CUDA is not available")
diff --git a/test/registered/distributed/test_disaggregation_different_tp.py b/test/registered/distributed/test_disaggregation_different_tp.py
index 1bd1b5102..7ea78bfbf 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=600, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=330, 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 ba88a48d7..b76cf1b18 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=580, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=400, 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 87eb64f2f..edbb177eb 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=500, suite="stage-c-test-8-gpu-h200")
+register_cuda_ci(est_time=450, suite="stage-c-test-8-gpu-h200")
 
 
 @unittest.skipIf(is_in_ci(), "Temporarily disable the flaky test.")
diff --git a/test/registered/distributed/test_dp_attention.py b/test/registered/distributed/test_dp_attention.py
index 077a0327a..4767d4355 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=350, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=450, 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 3e1d65f74..57e3f24ce 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=350, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=230, 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_pp_single_node.py b/test/registered/distributed/test_pp_single_node.py
index 0dd5d4fe8..cfc26b638 100644
--- a/test/registered/distributed/test_pp_single_node.py
+++ b/test/registered/distributed/test_pp_single_node.py
@@ -31,7 +31,7 @@ from sglang.test.test_utils import (
     run_bench_one_batch_server,
 )
 
-register_cuda_ci(est_time=650, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=460, 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 28588b0fb..b44a8ab5c 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=181, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=110, 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 a8bb9b8a0..11c18bab8 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=563, suite="stage-c-test-deepep-8-gpu-h200")
+register_cuda_ci(est_time=490, 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 d89670528..01e92d0bd 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=531, suite="stage-c-test-deepep-4-gpu-h100")
+register_cuda_ci(est_time=430, 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 5ef70914c..37c618991 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=200, suite="stage-c-test-deepep-4-gpu-h100")
+register_cuda_ci(est_time=80, suite="stage-c-test-deepep-4-gpu-h100")
 
 ib_devices = get_rdma_devices_args()
 
diff --git a/test/registered/hicache/test_hicache_storage_mooncake_backend.py b/test/registered/hicache/test_hicache_storage_mooncake_backend.py
index 066f14cb8..5d928babd 100644
--- a/test/registered/hicache/test_hicache_storage_mooncake_backend.py
+++ b/test/registered/hicache/test_hicache_storage_mooncake_backend.py
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     is_in_ci,
 )
 
-register_cuda_ci(est_time=300, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=230, suite="stage-b-test-2-gpu-large")
 
 
 class HiCacheStorageMooncakeBackendBaseMixin(HiCacheStorageBaseMixin):
diff --git a/test/registered/hicache/test_hicache_storage_runtime_attach_detach.py b/test/registered/hicache/test_hicache_storage_runtime_attach_detach.py
index ea5a6a861..3c9f07224 100644
--- a/test/registered/hicache/test_hicache_storage_runtime_attach_detach.py
+++ b/test/registered/hicache/test_hicache_storage_runtime_attach_detach.py
@@ -28,7 +28,7 @@ from sglang.test.test_utils import (
 )
 from sglang.utils import wait_for_http_ready
 
-register_cuda_ci(est_time=200, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=130, suite="stage-b-test-2-gpu-large")
 
 
 class TestHiCacheStorageRuntimeAttachDetach(CustomTestCase):
diff --git a/test/registered/hicache/test_hicache_variants.py b/test/registered/hicache/test_hicache_variants.py
index 89c9006a1..956d61d41 100644
--- a/test/registered/hicache/test_hicache_variants.py
+++ b/test/registered/hicache/test_hicache_variants.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=524, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=410, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=524, suite="stage-b-test-1-gpu-small-amd")
 """
 Consolidated HiCache variant tests.
diff --git a/test/registered/mla/test_mla_int8_deepseek_v3.py b/test/registered/mla/test_mla_int8_deepseek_v3.py
index 1faf4846e..1eb00dcce 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=341, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=290, suite="stage-b-test-1-gpu-large")
 
 
 class TestMLADeepseekV3ChannelInt8(CustomTestCase):
diff --git a/test/registered/models/test_nvidia_nemotron_3_nano.py b/test/registered/models/test_nvidia_nemotron_3_nano.py
index 1d26abd21..aba13bf7f 100644
--- a/test/registered/models/test_nvidia_nemotron_3_nano.py
+++ b/test/registered/models/test_nvidia_nemotron_3_nano.py
@@ -4,7 +4,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.lm_eval_kit import LMEvalMixin
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
-register_cuda_ci(est_time=660, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=540, suite="stage-b-test-2-gpu-large")
 
 NEMOTRON_3_NANO_THINKING_ARGS = [
     "--trust-remote-code",
diff --git a/test/registered/models/test_transformers_backend_eval.py b/test/registered/models/test_transformers_backend_eval.py
index 3b7f17364..16125aa57 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=180, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=40, 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 d14cde01d..39339f6de 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=450, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=260, 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/moe/test_torch_compile_moe.py b/test/registered/moe/test_torch_compile_moe.py
index e62af0cb9..998ddc61c 100644
--- a/test/registered/moe/test_torch_compile_moe.py
+++ b/test/registered/moe/test_torch_compile_moe.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=210, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=120, 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/openai_server/basic/test_anthropic_server.py b/test/registered/openai_server/basic/test_anthropic_server.py
index 2ea535f84..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=120, 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 6cfc3ee7e..31c21304c 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=120, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=50, 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/function_call/test_anthropic_tool_use.py b/test/registered/openai_server/function_call/test_anthropic_tool_use.py
index 1653b216d..7904b0b39 100644
--- a/test/registered/openai_server/function_call/test_anthropic_tool_use.py
+++ b/test/registered/openai_server/function_call/test_anthropic_tool_use.py
@@ -25,7 +25,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=50, suite="stage-b-test-1-gpu-large")
 register_amd_ci(est_time=140, suite="stage-b-test-1-gpu-small-amd")
 
 # System message to guide Llama3.2 to produce proper tool call format
diff --git a/test/registered/perf/test_bench_serving_1gpu_part1.py b/test/registered/perf/test_bench_serving_1gpu_part1.py
index a6ca1d336..9e2778c57 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=1000, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=1140, 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_vlm_perf_5090.py b/test/registered/perf/test_vlm_perf_5090.py
index 8f7d4cd04..8802c75b3 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=600, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=410, 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/prefill_only/test_embedding_models.py b/test/registered/prefill_only/test_embedding_models.py
index 3060ae304..bc8f48059 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=200, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=90, 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 115b26b2c..64d9c5dba 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=270, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=310, 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/quant/test_awq.py b/test/registered/quant/test_awq.py
index 06c860a63..c2a267652 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=700, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=260, 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_deepseek_v32_fp4_4gpu.py b/test/registered/quant/test_deepseek_v32_fp4_4gpu.py
index 19b0a1dba..f618d8332 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=1200, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=530, 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 32eeb9e9b..db2298423 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=1200, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=610, 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 f54149952..3a9030cfd 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=1200, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=450, 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_fp8_blockwise_gemm.py b/test/registered/quant/test_fp8_blockwise_gemm.py
index 30a04a1fc..a9602d94c 100644
--- a/test/registered/quant/test_fp8_blockwise_gemm.py
+++ b/test/registered/quant/test_fp8_blockwise_gemm.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=420, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=400, suite="stage-c-test-4-gpu-b200")
 
 MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507-FP8"
 MXFP8_MODEL_PATH = "zianglih/Qwen3-4B-Instruct-2507-MXFP8"
diff --git a/test/registered/quant/test_marlin_moe.py b/test/registered/quant/test_marlin_moe.py
index 310e56196..9a043882f 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=200, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=130, 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.py b/test/registered/quant/test_nvfp4_gemm.py
index f784973c9..dacbb04b9 100644
--- a/test/registered/quant/test_nvfp4_gemm.py
+++ b/test/registered/quant/test_nvfp4_gemm.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=360, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=300, suite="stage-c-test-4-gpu-b200")
 
 MODEL_PATH = "nvidia/Llama-3.1-8B-Instruct-NVFP4"
 
diff --git a/test/registered/quant/test_quantization.py b/test/registered/quant/test_quantization.py
index cdf0b0e61..083282b87 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=370, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=270, 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_torchao.py b/test/registered/quant/test_torchao.py
index a53b929a6..16dc187a7 100644
--- a/test/registered/quant/test_torchao.py
+++ b/test/registered/quant/test_torchao.py
@@ -5,7 +5,7 @@ import requests
 from sglang import Engine
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=200, 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=230, suite="stage-b-test-1-gpu-small-amd")
 from sglang.lang.chat_template import get_chat_template_by_model_path
 from sglang.srt.utils import kill_process_tree
diff --git a/test/registered/quant/test_w4a8_deepseek_v3.py b/test/registered/quant/test_w4a8_deepseek_v3.py
index f30e16d2a..72e717eb8 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=520, suite="stage-c-test-8-gpu-h20")
+register_cuda_ci(est_time=700, suite="stage-c-test-8-gpu-h20")
 
 
 class TestDeepseekV3W4afp8(CustomTestCase):
diff --git a/test/registered/rl/test_update_weights_from_disk_mxfp8.py b/test/registered/rl/test_update_weights_from_disk_mxfp8.py
index 504eb1f1d..a1111c138 100644
--- a/test/registered/rl/test_update_weights_from_disk_mxfp8.py
+++ b/test/registered/rl/test_update_weights_from_disk_mxfp8.py
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=200, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=100, suite="stage-c-test-4-gpu-b200")
 
 import unittest
 
diff --git a/test/registered/sampling/test_pytorch_sampling_backend.py b/test/registered/sampling/test_pytorch_sampling_backend.py
index 6abea3c3e..3b6df45f9 100644
--- a/test/registered/sampling/test_pytorch_sampling_backend.py
+++ b/test/registered/sampling/test_pytorch_sampling_backend.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=150, 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_chunked_prefill.py b/test/registered/scheduler/test_chunked_prefill.py
index bc22976c3..a4257dedb 100644
--- a/test/registered/scheduler/test_chunked_prefill.py
+++ b/test/registered/scheduler/test_chunked_prefill.py
@@ -7,7 +7,7 @@ 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, run_mulit_request_test
 
-register_cuda_ci(est_time=550, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=360, suite="stage-b-test-1-gpu-small")
 register_amd_ci(est_time=312, 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 0628c97e4..42735b245 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=550, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=470, 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/spec/eagle/test_deepseek_v3_fp4_mtp_small.py b/test/registered/spec/eagle/test_deepseek_v3_fp4_mtp_small.py
index d1bce0c96..fe589679f 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=900, suite="stage-b-test-4-gpu-b200")
+register_cuda_ci(est_time=240, 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 4cf48afc1..2830228a3 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=200, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=160, 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_dp_attention.py b/test/registered/spec/eagle/test_eagle_dp_attention.py
index dc20b3981..fe04bdb10 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=200, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=100, 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 cecc2eb53..bce64a0c4 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=450, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=270, 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 7c726acfb..1585a4bcc 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=600, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=690, suite="stage-b-test-1-gpu-large")
 
 
 class TestEAGLEServerBasic(EagleServerBase):
diff --git a/test/registered/vlm/test_vision_openai_server_a.py b/test/registered/vlm/test_vision_openai_server_a.py
index ec4ab28df..f64f7b8bc 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=957, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=700, 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 de5369d80..44af1e5d8 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=447, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=620, 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 d62df9b0b..6f1a12394 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=200, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=140, suite="stage-c-test-4-gpu-h100")
 
 QWEN35_27B_MODEL = "Qwen/Qwen3.5-27B"
 MMMU_ACCURACY_THRESHOLD = 0.65