diff --git a/test/registered/metrics/test_priority_metrics.py b/test/registered/metrics/test_priority_metrics.py
index 8cbdd704d..1fc5cc37b 100644
--- a/test/registered/metrics/test_priority_metrics.py
+++ b/test/registered/metrics/test_priority_metrics.py
@@ -16,7 +16,11 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=60, suite="stage-b-test-small-1-gpu")
+register_cuda_ci(
+    est_time=60,
+    suite="stage-b-test-small-1-gpu",
+    disabled="Flaky: no TTFT histogram samples. See https://github.com/sgl-project/sglang/actions/runs/22787468608/job/66118130037",
+)
 register_amd_ci(est_time=60, suite="stage-b-test-small-1-gpu-amd")
 
 _MODEL_NAME = "Qwen/Qwen3-0.6B"
diff --git a/test/registered/sessions/test_session_latency.py b/test/registered/sessions/test_session_latency.py
index 63e1e7746..a8160e3f8 100644
--- a/test/registered/sessions/test_session_latency.py
+++ b/test/registered/sessions/test_session_latency.py
@@ -35,7 +35,11 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=100, suite="stage-b-test-large-1-gpu")
+register_cuda_ci(
+    est_time=100,
+    suite="stage-b-test-large-1-gpu",
+    disabled="Flaky: streaming vs regular session 1/300 turns differ. See https://github.com/sgl-project/sglang/actions/runs/22790998325/job/66117795513",
+)
 
 NUM_TURNS = 300
 INPUT_LEN = 16