diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml
index f8cc486b2..9d60f5968 100644
--- a/.github/configs/nvidia-master.yaml
+++ b/.github/configs/nvidia-master.yaml
@@ -3002,7 +3002,7 @@ dsv4-fp8-h200-sglang-mtp:
   # field, so dp-attn=true is used as the existing vLLM script switch for DP4
   # layouts on 4 allocated GPUs.
 dsv4-fp4-b300-vllm:
-  image: vllm/vllm-openai:v0.21.0
+  image: vllm/vllm-openai:v0.22.0
   model: deepseek-ai/DeepSeek-V4-Pro
   model-prefix: dsv4
   runner: b300
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 935cded22..db776b4ea 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -3200,3 +3200,9 @@
     - "Bump image to lmsysorg/sglang-rocm:v0.5.12.post1-rocm720-mi35x-20260523, 1P1D TP8/EP1, dp-attn false, conc [8..512]"
     - "MoRI conn.py overlay (48e459bd) via job.slurm; launcher qwen3.5_fp4_mi355x_sglang-disagg.sh"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1579
+
+- config-keys:
+    - dsv4-fp4-b300-vllm
+  description:
+    - "Update DSv4 FP4 B300 vLLM image tag to v0.22.0"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1588