diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml
index e63a79eaf..10fff4091 100644
--- a/.github/configs/nvidia-master.yaml
+++ b/.github/configs/nvidia-master.yaml
@@ -4427,7 +4427,7 @@ minimaxm2.5-fp4-b200-vllm-agentic:
   # MiniMax-M2.5 FP4 B200 vLLM recipe as-is until B300-specific tuning is available.
 
 minimaxm2.5-fp4-b300-vllm:
-  image: vllm/vllm-openai:v0.19.0-cu130
+  image: vllm/vllm-openai:v0.21.0
   model: nvidia/MiniMax-M2.5-NVFP4
   model-prefix: minimaxm2.5
   runner: b300
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index e8db0c360..20759de79 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -2808,3 +2808,9 @@
   description:
     - "Update vLLM image from v0.20.0-cu130 (18d old) to v0.21.0"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1456
+
+- config-keys:
+    - minimaxm2.5-fp4-b300-vllm
+  description:
+    - "Update vLLM image from v0.19.0-cu130 (26d old) to v0.21.0"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1453