diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml
index 8b031a5dc..a5cf9b5f7 100644
--- a/.github/configs/nvidia-master.yaml
+++ b/.github/configs/nvidia-master.yaml
@@ -2500,7 +2500,7 @@ kimik2.5-int4-h200-vllm:
       - { tp: 8, conc-start: 4, conc-end: 64 }
 
 kimik2.5-fp4-b200-vllm:
-  image: vllm/vllm-openai:v0.17.0
+  image: vllm/vllm-openai:v0.20.2
   model: nvidia/Kimi-K2.5-NVFP4
   model-prefix: kimik2.5
   runner: b200
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index c599199c9..6ee979a4f 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -2329,6 +2329,12 @@
     - "Add Qwen3.5-397B-A17B FP8 MI355X ATOM benchmark configs with and without MTP"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1310
 
+- config-keys:
+    - kimik2.5-fp4-b200-vllm
+  description:
+    - "Update vLLM image from v0.17.0 to v0.20.2"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1336
+
 - config-keys:
     - gptoss-fp4-h100-vllm
   description: