diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index f10d232137..a60792bd68 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -7234,7 +7234,7 @@ minimaxm3-fp4-b300-vllm-agentic-mtp:
 # the same 3 TB AgentX ceiling before the proportional-GPU rule is applied.
 # GPU-resident points receive a zero budget.
 minimaxm3-fp4-b200-vllm-agentic-mtp:
-  image: vllm/vllm-openai:nightly-5e35a6f4f9bbc217c599692157ca985c894373f7
+  image: vllm/vllm-openai:nightly-ac7509e2b1db40fec2f03dde1ed4e9dfdc2338c9
   model: nvidia/MiniMax-M3-NVFP4
   model-prefix: minimaxm3
   runner: cluster:b200-dgxc
@@ -7245,8 +7245,11 @@ minimaxm3-fp4-b200-vllm-agentic-mtp:
     agentic-coding:
     - dram-utilization: 0.683
       search-space:
-      - { tp: 4, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 2, 5, 8, 10, 12, 15, 20] }
-      - { tp: 4, spec-decoding: mtp, kv-offloading: dram, kv-offload-backend: { name: vllm-simple }, conc-list: [20, 30, 40] }
+      - { tp: 8, spec-decoding: mtp, kv-offloading: none, conc-list: [1] }
+      - { tp: 4, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 5, 10, 15, 20] }
+    - dram-utilization: 1.0
+      search-space:
+      - { tp: 4, spec-decoding: mtp, kv-offloading: dram, kv-offload-backend: { name: vllm-simple }, conc-list: [15, 20, 25, 30, 32, 34, 36, 38, 40] }
 # Preserve the B200 TP4 search space and add the GB200 Pareto candidates found
 # by direct DEP and P/D tuning.
 minimaxm3-fp4-gb200-dynamo-vllm-agentic-agg-mtp:
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 24cad03446..889d4db484 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5999,3 +5999,11 @@
     - "Add the tuned GB200 MiniMax-M3 FP4 AgentX frontier with EAGLE3, the B200 TP4 baseline, DEP4/DEP8, SimpleCPU offload, and KV-routed P/D."
     - "Use full-decode-only CUDA graphs for TP4 stability."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2609
+
+- config-keys:
+    - minimaxm3-fp4-b200-vllm-agentic-mtp
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Refresh the MiniMax-M3 B200 AgentX submission: no-offload TP8 C1 and TP4 C1/C5/C10/C15/C20, plus a TP4 SimpleCPU KV-offload sweep at C15–C40 using the full-capacity DRAM budget."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2611