diff --git a/benchmarks/single_node/agentic/minimaxm3_fp4_mi355x_mtp.sh b/benchmarks/single_node/agentic/minimaxm3_fp4_mi355x_mtp.sh
new file mode 100644
index 000000000..3239327e5
--- /dev/null
+++ b/benchmarks/single_node/agentic/minimaxm3_fp4_mi355x_mtp.sh
@@ -0,0 +1,154 @@
+#!/usr/bin/env bash
+set -euo pipefail
+set -x
+
+# Agentic trace replay benchmark for Minimax-M3 FP4 on MI355X using vLLM.
+#
+# Required env vars:
+#   MODEL, MODEL_PATH, TP, CONC, KV_OFFLOADING, KV_OFFLOAD_BACKEND,
+#   TOTAL_CPU_DRAM_GB, RESULT_DIR, DURATION, EP_SIZE, DP_ATTENTION
+
+source "$(dirname "$0")/../../benchmark_lib.sh"
+
+# Force the eval framework to lm-eval for this recipe. run_eval derives its
+# default as swebench for agentic scenarios (scenario_default=swebench when
+# IS_AGENTIC/SCENARIO_TYPE=agentic-coding), but EVAL_FRAMEWORK takes precedence
+# over that default (benchmark_lib.sh: framework=${EVAL_FRAMEWORK:-...}), so
+# setting it here makes the effective framework always lm-eval, never swebench.
+export EVAL_FRAMEWORK="lm-eval"
+
+check_env_vars MODEL TP CONC KV_OFFLOADING KV_OFFLOAD_BACKEND TOTAL_CPU_DRAM_GB RESULT_DIR DURATION EP_SIZE DP_ATTENTION
+
+echo "MODEL=$MODEL TP=$TP CONC=$CONC KV_OFFLOADING=$KV_OFFLOADING TOTAL_CPU_DRAM_GB=$TOTAL_CPU_DRAM_GB RESULT_DIR=$RESULT_DIR DURATION=$DURATION EP_SIZE=$EP_SIZE DP_ATTENTION=$DP_ATTENTION"
+
+PORT=8888
+
+if [[ -n "${SLURM_JOB_ID+x}" ]]; then
+    echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+# ROCR/HIP visibility for vLLM 0.14+
+if [[ -n "${ROCR_VISIBLE_DEVICES+x}" ]]; then
+    export HIP_VISIBLE_DEVICES="$ROCR_VISIBLE_DEVICES"
+fi
+
+if [[ -n "${MODEL_PATH:-}" ]]; then
+    if [[ ! -d "$MODEL_PATH" || -z "$(ls -A "$MODEL_PATH" 2>/dev/null)" ]]; then
+        hf download "$MODEL" --local-dir "$MODEL_PATH"
+    fi
+else
+    hf download "$MODEL"
+    export MODEL_PATH="$MODEL"
+fi
+
+rocm-smi || true
+amd-smi || true
+
+resolve_trace_source
+install_agentic_deps
+
+DRAFT_MODEL="Inferact/MiniMax-M3-EAGLE3"
+NUM_SPEC_TOKENS=5
+# Throughput pins synthetic EAGLE3 acceptance to the minimax-m3 golden AL
+# (thinking_on, num_speculative_tokens=5, golden_al_distribution/minimaxm3_eagle3.yaml).
+# MiniMax-M3 is an interleaved-thinking model and the replay runs it with
+# thinking active, so thinking_on is the matching curve. The EVAL_ONLY accuracy
+# run uses real target verification instead -- synthetic acceptance bypasses
+# verification and corrupts the eval score.
+SYNTHETIC_ACCEPT_LEN=3.35
+if [ "${EVAL_ONLY:-false}" = "true" ]; then
+    SPEC_CONFIG="{\"method\": \"eagle3\", \"model\": \"$DRAFT_MODEL\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS, \"attention_backend\": \"TRITON_ATTN\"}"
+else
+    SPEC_CONFIG="{\"method\": \"eagle3\", \"model\": \"$DRAFT_MODEL\", \"num_speculative_tokens\": $NUM_SPEC_TOKENS, \"attention_backend\": \"TRITON_ATTN\", \"rejection_sample_method\": \"synthetic\", \"synthetic_acceptance_length\": $SYNTHETIC_ACCEPT_LEN}"
+fi
+
+# ---- Server config ----------------------------------------------------------
+SERVER_LOG="$RESULT_DIR/server.log"
+LMCACHE_LOG="$RESULT_DIR/lmcache_server.log"
+mkdir -p "$RESULT_DIR"
+
+OFFLOAD_ARGS=(--no-enable-prefix-caching)
+
+case "$KV_OFFLOAD_BACKEND" in
+    vllm-simple)
+        unset VLLM_USE_SIMPLE_KV_OFFLOAD
+        # Use vLLM's regular native KV-offload path (OffloadingConnector),
+        # NOT the SimpleCPUOffloadConnector. The "native" backend resolves to
+        # OffloadingConnector by default; setting VLLM_USE_SIMPLE_KV_OFFLOAD=1
+        # would switch it to SimpleCPUOffloadConnector. We intentionally leave
+        # that env var UNSET here so the regular OffloadingConnector path is
+        # used. The shortcut --kv_offloading_backend native + --kv_offloading_size
+        # form constructs the KVTransferConfig at engine startup
+        # (vllm/config/vllm.py:662).
+
+        # Remove --disable-hybrid-kv-cache-manager and enable hybrid kv cache manager (default)
+        # This gives extra cache hit than disabling hybrid kv cache manager
+        OFFLOAD_ARGS=(
+            --kv_offloading_backend native
+            --kv_offloading_size "$TOTAL_CPU_DRAM_GB"
+        )
+        ;;
+esac
+
+# ---- LLM server config ----------------------------------------------------------
+PARALLEL_ARGS=(--tensor-parallel-size "$TP")
+if [ "${DP_ATTENTION}" = "true" ]; then
+    PARALLEL_ARGS=(
+        --tensor-parallel-size 1
+        --data-parallel-size "$TP"
+        --enable-expert-parallel
+    )
+elif [ "$EP_SIZE" -gt 1 ]; then
+    PARALLEL_ARGS+=(--enable-expert-parallel)
+fi
+
+echo "Starting vllm server..."
+export PYTHONNOUSERSITE=1
+
+export VLLM_ENGINE_READY_TIMEOUT_S=3600
+export VLLM_USE_BREAKABLE_CUDAGRAPH=0
+export VLLM_ROCM_USE_AITER=1
+export VLLM_ROCM_USE_AITER_MOE=1
+export VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS=1
+# INT4 quantized all-reduce for the (~1.5 MB) decode all-reduces, which are the
+# single biggest decode kernel at high concurrency. The MIN_SIZE_KB override is
+# required: vLLM's default INT4 quick-reduce size gate for (bf16, TP4) is 16 MB,
+# so it never fires for decode-sized tensors without it.
+export VLLM_ROCM_QUICK_REDUCE_QUANTIZATION=INT4
+export VLLM_ROCM_QUICK_REDUCE_CAST_BF16_TO_FP16=0
+export VLLM_ROCM_QUICK_REDUCE_QUANTIZATION_MIN_SIZE_KB=256
+
+VLLM_CMD=(
+    vllm serve "$MODEL_PATH"
+    --served-model-name "$MODEL"
+    --host 0.0.0.0
+    --port "$PORT"
+    "${PARALLEL_ARGS[@]}"
+    --speculative-config "$SPEC_CONFIG"
+    --trust-remote-code
+    --block-size 128
+    --gpu-memory-utilization 0.80
+    --language-model-only
+    --attention-backend TRITON_ATTN
+    --moe-backend aiter
+    --kv-cache-dtype fp8
+    --tool-call-parser minimax_m3
+    --enable-auto-tool-choice
+    --max-num-seqs "$CONC"
+    "${OFFLOAD_ARGS[@]}"
+)
+printf '%q ' "${VLLM_CMD[@]}" | tee "$RESULT_DIR/vllm_command.txt"
+printf '\n' | tee -a "$RESULT_DIR/vllm_command.txt"
+"${VLLM_CMD[@]}" > "$SERVER_LOG" 2>&1 &
+SERVER_PID=$!
+echo "Server PID: $SERVER_PID"
+
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+# ---- Run benchmark ----------------------------------------------------------
+if [ "${EVAL_ONLY}" = "true" ]; then
+    run_eval --port "$PORT"
+else
+    build_replay_cmd "$RESULT_DIR"
+    run_agentic_replay_and_write_outputs "$RESULT_DIR"
+fi
diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index 0d46afe2e..ab1fcf64a 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -2361,3 +2361,17 @@ dsv4-fp8-mi325x-vllm-mtp:
       # is 55.8%. conc128 passed cleanly on the memory-tightest SKU (MI300X);
       # cap 8k1k MTP at 128 (normal holds 256, 1k1k holds 512).
       - { tp: 8, conc-start: 4, conc-end: 128, spec-decoding: mtp }
+
+minimaxm3-fp4-mi355x-vllm-agentic-mtp:
+  image: vllm/vllm-openai-rocm:nightly-cb8104839c141609d99f1254459ef3a4f1bd4263
+  model: amd/MiniMax-M3-MXFP4
+  model-prefix: minimaxm3
+  runner: cluster:mi355x-amds
+  precision: fp4
+  framework: vllm
+  multinode: false
+  scenarios:
+    agentic-coding:
+    - dram-utilization: 0.80
+      search-space:
+      - { tp: 4, kv-offloading: dram, kv-offload-backend: { name: vllm-simple }, conc-list: [1, 4, 8, 12, 16], spec-decoding: mtp }
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 41bd57494..6b50f2460 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5440,3 +5440,11 @@
     - "Add MiniMax-M3-MXFP8 GB300 8k1k disaggregated Dynamo-vLLM EAGLE3 MTP Pareto on vLLM nightly 5e35a6f4f"
     - "The six dep8 decode configurations enable the CUTLASS MSA sparse decode backend and FULL_DECODE_ONLY cudagraph mode; the five low-concurrency c1/c4/c8 configurations use neither"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2486
+
+- config-keys:
+    - minimaxm3-fp4-mi355x-vllm-agentic-mtp
+  description:
+    - "Add Minimax-M3 FP4 vLLM Single Node Agentic Support"
+    - "Image: vllm/vllm-openai-rocm:nightly-cb8104839c141609d99f1254459ef3a4f1bd4263"
+    - "Throughput runs use synthetic EAGLE3 acceptance pinned to the minimax-m3 golden AL (thinking_on, num_speculative_tokens=5, AL=3.35); EVAL_ONLY accuracy runs keep real target verification"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2487