diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml
index a255a9d77f..aaa88c634f 100644
--- a/.github/configs/nvidia-master.yaml
+++ b/.github/configs/nvidia-master.yaml
@@ -12035,6 +12035,222 @@ minimaxm3-fp8-b300-dynamo-vllm:
           ep: 8
           dp-attn: false
 
+# MiniMax-M3 NVFP4 disagg sweep on the same B300 topology matrix as the MXFP8
+# baseline above. The image includes vLLM PR #46380, so no runtime patch is
+# needed.
+minimaxm3-fp4-b300-dynamo-vllm:
+  image: vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41
+  model: nvidia/MiniMax-M3-NVFP4
+  model-prefix: minimaxm3
+  runner: b300
+  precision: fp4
+  framework: dynamo-vllm
+  multinode: true
+  disagg: true
+  scenarios:
+    fixed-seq-len:
+    - isl: 1024
+      osl: 1024
+      search-space:
+      - conc-list: [4, 16, 64, 128, 4096]
+        prefill:
+          num-worker: 1
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/1k1k/1p1d-dep2-tep8-1k1k.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: false
+      - conc-list: [1, 4, 8, 16]
+        prefill:
+          num-worker: 1
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/1k1k/1p1d-dep2-tp4-1k1k.yaml"
+        decode:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - conc-list: [2048]
+        prefill:
+          num-worker: 1
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/1k1k/1p2d-dep2-dep4-1k1k.yaml"
+        decode:
+          num-worker: 2
+          tp: 4
+          ep: 4
+          dp-attn: true
+      - conc-list: [512, 4096]
+        prefill:
+          num-worker: 2
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/1k1k/2p1d-dep2-dep8-1k1k.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: true
+      - conc-list: [32]
+        prefill:
+          num-worker: 2
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/1k1k/2p1d-dep2-tep8-1k1k.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: false
+      - conc-list: [16]
+        prefill:
+          num-worker: 2
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/1k1k/2p2d-dep2-tep8-1k1k.yaml"
+        decode:
+          num-worker: 2
+          tp: 8
+          ep: 8
+          dp-attn: false
+      - conc-list: [4]
+        prefill:
+          num-worker: 3
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/1k1k/3p2d-dep2-tep8-1k1k.yaml"
+        decode:
+          num-worker: 2
+          tp: 8
+          ep: 8
+          dp-attn: false
+    - isl: 8192
+      osl: 1024
+      search-space:
+      - conc-list: [256, 512]
+        prefill:
+          num-worker: 2
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-dep8-8k1k.yaml"
+        decode:
+          num-worker: 2
+          tp: 8
+          ep: 8
+          dp-attn: true
+      - conc-list: [16]
+        prefill:
+          num-worker: 2
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-tep8-8k1k.yaml"
+        decode:
+          num-worker: 2
+          tp: 8
+          ep: 8
+          dp-attn: false
+      - conc-list: [4096]
+        prefill:
+          num-worker: 4
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-dep8-8k1k.yaml"
+        decode:
+          num-worker: 2
+          tp: 8
+          ep: 8
+          dp-attn: true
+      - conc-list: [1, 4, 8, 16]
+        prefill:
+          num-worker: 1
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-dep2-tp4-8k1k.yaml"
+        decode:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - conc-list: [4096]
+        prefill:
+          num-worker: 4
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-tep4-8k1k.yaml"
+        decode:
+          num-worker: 2
+          tp: 4
+          ep: 4
+          dp-attn: false
+      - conc-list: [16, 32, 64, 128]
+        prefill:
+          num-worker: 1
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep4-8k1k.yaml"
+        decode:
+          num-worker: 4
+          tp: 4
+          ep: 4
+          dp-attn: false
+      - conc-list: [16]
+        prefill:
+          num-worker: 1
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p2d-dep2-tep4-8k1k.yaml"
+        decode:
+          num-worker: 2
+          tp: 4
+          ep: 4
+          dp-attn: false
+      - conc-list: [4]
+        prefill:
+          num-worker: 1
+          tp: 2
+          ep: 2
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep8-8k1k.yaml"
+        decode:
+          num-worker: 4
+          tp: 8
+          ep: 8
+          dp-attn: false
+
 # MiniMax-M3 GB300 disagg sweep — refreshed recipe set (no Marlin variants).
 # All prefill DEP2 (TP1 DP2 EP, 2 GPU/worker). Decode: DEP4, TEP8, DEP8, TEP4.
 # 4 GPU/node (GB300 NVL72). kv-cache-dtype=fp8. srun_options mem=0 required.
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/1p1d-dep2-tep8-1k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/1p1d-dep2-tep8-1k1k.yaml
new file mode 100644
index 0000000000..486af05573
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/1p1d-dep2-tep8-1k1k.yaml
@@ -0,0 +1,81 @@
+name: "minimax-m3-vllm-disagg-b300-1p1d-fp4-dep2-tep8-1k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 1
+  prefill_workers: 1
+  decode_workers: 1
+  gpus_per_prefill: 2
+  gpus_per_decode: 8
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  decode_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 2048
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 8
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 4096
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 8192
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  concurrencies: "4x16x64x128x4096"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/1p1d-dep2-tp4-1k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/1p1d-dep2-tp4-1k1k.yaml
new file mode 100644
index 0000000000..532b78a103
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/1p1d-dep2-tp4-1k1k.yaml
@@ -0,0 +1,82 @@
+name: "minimax-m3-vllm-disagg-b300-1p1d-fp4-dep2-tp4-1k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 1
+  prefill_workers: 1
+  decode_workers: 1
+  gpus_per_prefill: 2
+  gpus_per_decode: 4
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+  allow_prefill_decode_colocation: true
+
+  prefill_environment:
+    UCX_TLS: "cuda_ipc,cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  decode_environment:
+    UCX_TLS: "cuda_ipc,cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 2048
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 4
+      enable-expert-parallel: false
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 4096
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 2048
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  concurrencies: "1x4x8x16"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/1p2d-dep2-dep4-1k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/1p2d-dep2-dep4-1k1k.yaml
new file mode 100644
index 0000000000..fde8442a18
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/1p2d-dep2-dep4-1k1k.yaml
@@ -0,0 +1,83 @@
+name: "minimax-m3-vllm-disagg-b300-1p2d-fp4-dep2-dep4-1k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 1
+  prefill_workers: 1
+  decode_workers: 2
+  gpus_per_prefill: 2
+  gpus_per_decode: 4
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  decode_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 2048
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 4
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 1024
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 8192
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  concurrencies: "2048"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/2p1d-dep2-dep8-1k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/2p1d-dep2-dep8-1k1k.yaml
new file mode 100644
index 0000000000..ed3b5f9950
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/2p1d-dep2-dep8-1k1k.yaml
@@ -0,0 +1,83 @@
+name: "minimax-m3-vllm-disagg-b300-2p1d-fp4-dep2-dep8-1k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 1
+  prefill_workers: 2
+  decode_workers: 1
+  gpus_per_prefill: 2
+  gpus_per_decode: 8
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  decode_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 2048
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 8
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 4096
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 8192
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  concurrencies: "512x4096"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/2p1d-dep2-tep8-1k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/2p1d-dep2-tep8-1k1k.yaml
new file mode 100644
index 0000000000..0784283b91
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/2p1d-dep2-tep8-1k1k.yaml
@@ -0,0 +1,81 @@
+name: "minimax-m3-vllm-disagg-b300-2p1d-fp4-dep2-tep8-1k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 1
+  prefill_workers: 2
+  decode_workers: 1
+  gpus_per_prefill: 2
+  gpus_per_decode: 8
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  decode_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 2048
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 8
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 4096
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 8192
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  concurrencies: "32"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/2p2d-dep2-tep8-1k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/2p2d-dep2-tep8-1k1k.yaml
new file mode 100644
index 0000000000..59c52da00c
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/2p2d-dep2-tep8-1k1k.yaml
@@ -0,0 +1,81 @@
+name: "minimax-m3-vllm-disagg-b300-2p2d-fp4-dep2-tep8-1k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 2
+  prefill_workers: 2
+  decode_workers: 2
+  gpus_per_prefill: 2
+  gpus_per_decode: 8
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  decode_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 2048
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 8
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 4096
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 8192
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  concurrencies: "16"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/3p2d-dep2-tep8-1k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/3p2d-dep2-tep8-1k1k.yaml
new file mode 100644
index 0000000000..7e9f7dec31
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/1k1k/3p2d-dep2-tep8-1k1k.yaml
@@ -0,0 +1,81 @@
+name: "minimax-m3-vllm-disagg-b300-3p2d-fp4-dep2-tep8-1k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 2
+  prefill_workers: 3
+  decode_workers: 2
+  gpus_per_prefill: 2
+  gpus_per_decode: 8
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  decode_environment:
+    UCX_TLS: "cuda_copy,rc"
+    VLLM_FLOAT32_MATMUL_PRECISION: "high"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 2048
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 8
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 2304
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 4096
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 8192
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  concurrencies: "4"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-dep2-tp4-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-dep2-tp4-8k1k.yaml
new file mode 100644
index 0000000000..be2683d0ca
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-dep2-tp4-8k1k.yaml
@@ -0,0 +1,86 @@
+name: "minimax-m3-vllm-disagg-b300-1p1d-fp4-dep2-tp4-8k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 1
+  prefill_workers: 1
+  decode_workers: 1
+  gpus_per_prefill: 2
+  gpus_per_decode: 4
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+  allow_prefill_decode_colocation: true
+
+  prefill_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_ipc,cuda_copy,rc"
+
+  decode_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_ipc,cuda_copy,rc"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 16384
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 4
+      enable-expert-parallel: false
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 1024
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 2048
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  concurrencies: "1x4x8x16"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p2d-dep2-tep4-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p2d-dep2-tep4-8k1k.yaml
new file mode 100644
index 0000000000..5be198f113
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p2d-dep2-tep4-8k1k.yaml
@@ -0,0 +1,85 @@
+name: "minimax-m3-vllm-disagg-b300-1p2d-fp4-dep2-tep4-8k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 1
+  prefill_workers: 1
+  decode_workers: 2
+  gpus_per_prefill: 2
+  gpus_per_decode: 4
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 16384
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 4
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 512
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 4096
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  concurrencies: "16"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep4-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep4-8k1k.yaml
new file mode 100644
index 0000000000..90d688f615
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep4-8k1k.yaml
@@ -0,0 +1,85 @@
+name: "minimax-m3-vllm-disagg-b300-1p4d-fp4-dep2-tep4-8k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 2
+  prefill_workers: 1
+  decode_workers: 4
+  gpus_per_prefill: 2
+  gpus_per_decode: 4
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 16384
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 4
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 512
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 4096
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  concurrencies: "16x32x64x128"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep8-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep8-8k1k.yaml
new file mode 100644
index 0000000000..2154742821
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep8-8k1k.yaml
@@ -0,0 +1,85 @@
+name: "minimax-m3-vllm-disagg-b300-1p4d-fp4-dep2-tep8-8k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 4
+  prefill_workers: 1
+  decode_workers: 4
+  gpus_per_prefill: 2
+  gpus_per_decode: 8
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 16384
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 8
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 1024
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 4096
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  concurrencies: "4"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-dep8-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-dep8-8k1k.yaml
new file mode 100644
index 0000000000..c49fd1ccbf
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-dep8-8k1k.yaml
@@ -0,0 +1,87 @@
+name: "minimax-m3-vllm-disagg-b300-2p2d-fp4-dep2-dep8-8k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 2
+  prefill_workers: 2
+  decode_workers: 2
+  gpus_per_prefill: 2
+  gpus_per_decode: 8
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 16384
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 8
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 1024
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 4096
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  concurrencies: "256x512"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-tep8-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-tep8-8k1k.yaml
new file mode 100644
index 0000000000..1b8dfd627f
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-tep8-8k1k.yaml
@@ -0,0 +1,85 @@
+name: "minimax-m3-vllm-disagg-b300-2p2d-fp4-dep2-tep8-8k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 2
+  prefill_workers: 2
+  decode_workers: 2
+  gpus_per_prefill: 2
+  gpus_per_decode: 8
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 16384
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 8
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 1024
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 4096
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  concurrencies: "16"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-dep8-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-dep8-8k1k.yaml
new file mode 100644
index 0000000000..73473aac94
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-dep8-8k1k.yaml
@@ -0,0 +1,87 @@
+name: "minimax-m3-vllm-disagg-b300-4p2d-fp4-dep2-dep8-8k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 2
+  prefill_workers: 4
+  decode_workers: 2
+  gpus_per_prefill: 2
+  gpus_per_decode: 8
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 16384
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 8
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 1024  # Per DP rank: 2 workers x DP8 = 16 ranks.
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 4096
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  concurrencies: "4096"
+  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-tep4-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-tep4-8k1k.yaml
new file mode 100644
index 0000000000..23c99d3282
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-tep4-8k1k.yaml
@@ -0,0 +1,85 @@
+name: "minimax-m3-vllm-disagg-b300-4p2d-fp4-dep2-tep4-8k1k"
+
+model:
+  path: "nvidia/MiniMax-M3-NVFP4"
+  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  decode_nodes: 1
+  prefill_workers: 4
+  decode_workers: 2
+  gpus_per_prefill: 2
+  gpus_per_decode: 4
+
+dynamo:
+  install: true
+  version: 1.3.0.dev20260614
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: false
+
+backend:
+  type: vllm
+  connector: null
+
+  prefill_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    VLLM_FLOAT32_MATMUL_PRECISION: high
+    UCX_TLS: "cuda_copy,rc"
+
+  vllm_config:
+    prefill:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 1
+      data-parallel-size: 2
+      data-parallel-rpc-port: 13345
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-cudagraph-capture-size: 2048
+      max-num-batched-tokens: 16384
+
+    decode:
+      no-enable-flashinfer-autotune: true
+      tensor-parallel-size: 4
+      enable-expert-parallel: true
+      trust-remote-code: true
+      no-enable-prefix-caching: true
+      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
+      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
+      kv-cache-dtype: fp8
+      block-size: 128
+      gpu-memory-utilization: 0.90
+      max-model-len: 9472
+      language-model-only: true
+      stream-interval: 32
+      max-num-seqs: 1024
+      max-num-batched-tokens: 16384
+      max-cudagraph-capture-size: 4096
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  concurrencies: "4096"
+  req_rate: "inf"
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index a112c63498..c318f2a2ac 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -4335,3 +4335,11 @@
     - "minimaxm3-fp8-mi355x-vllm-mtp: 1k1k sweeps TP8 (conc 4-32), TP8/EP8 (conc 1-256), TP4 (conc 1-2 and 32-64), TP4/EP4 (conc 128-256); 8k1k sweeps TP8 (conc 1 and 4-16), TP4 (conc 16-128)."
     - "Serving flags are otherwise unchanged."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1946
+
+- config-keys:
+    - minimaxm3-fp4-b300-dynamo-vllm
+  description:
+    - "Add MiniMax-M3 NVFP4 B300 disaggregated vLLM benchmarks via Dynamo for 1k1k and 8k1k STP (no MTP)"
+    - "Use nvidia/MiniMax-M3-NVFP4 from /scratch/models/MiniMax-M3-NVFP4 with vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41, which includes vllm-project/vllm PR #46380; no runtime patch needed"
+    - "Reuse the existing MXFP8 B300 topology and concurrency matrix across 15 srt-slurm recipes, while dropping the FP8-only Marlin override from TP4 decode"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1931
diff --git a/runners/launch_b300-nv.sh b/runners/launch_b300-nv.sh
index 7a7008691a..9f6cf1b07c 100644
--- a/runners/launch_b300-nv.sh
+++ b/runners/launch_b300-nv.sh
@@ -47,11 +47,14 @@ elif [[ $MODEL_PREFIX == "minimaxm2.5" && $PRECISION == "fp4" && $FRAMEWORK == "
 elif [[ $MODEL_PREFIX == "minimaxm2.5" && $PRECISION == "fp8" && $FRAMEWORK == "dynamo-vllm" ]]; then
     export MODEL_PATH="/data/models/MiniMax-M2.5"
     export SRT_SLURM_MODEL_PREFIX="minimax-m2.5-fp8"
+elif [[ $MODEL_PREFIX == "minimaxm3" && $PRECISION == "fp4" && $FRAMEWORK == "dynamo-vllm" ]]; then
+    export MODEL_PATH="/scratch/models/MiniMax-M3-NVFP4"
+    export SRT_SLURM_MODEL_PREFIX="nvidia/MiniMax-M3-NVFP4"
 elif [[ $MODEL_PREFIX == "minimaxm3" && $PRECISION == "fp8" && $FRAMEWORK == "dynamo-vllm" ]]; then
     export MODEL_PATH="/data/models/MiniMax-M3-MXFP8"
     export SRT_SLURM_MODEL_PREFIX="MiniMaxAI/MiniMax-M3-MXFP8"
 else
-    echo "Unsupported model: $MODEL_PREFIX-$PRECISION. Supported models are: dsr1-fp4, dsr1-fp8, dsv4-fp4 with dynamo-vllm, minimaxm2.5-fp4 with dynamo-vllm, minimaxm2.5-fp8 with dynamo-vllm, minimaxm3-fp8 with dynamo-vllm"
+    echo "Unsupported model: $MODEL_PREFIX-$PRECISION. Supported models are: dsr1-fp4, dsr1-fp8, dsv4-fp4 with dynamo-vllm, minimaxm2.5-fp4 with dynamo-vllm, minimaxm2.5-fp8 with dynamo-vllm, minimaxm3-fp4 with dynamo-vllm, minimaxm3-fp8 with dynamo-vllm"
     exit 1
 fi
 
@@ -85,18 +88,22 @@ elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "minimaxm2.5" && $PRECIS
     git checkout main
     mkdir -p recipes/vllm/minimax-m2.5-fp8
     cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m2.5-b300-fp8" recipes/vllm/minimax-m2.5-fp8
-elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "minimaxm3" && $PRECISION == "fp8" ]]; then
+elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "minimaxm3" && ( $PRECISION == "fp4" || $PRECISION == "fp8" ) ]]; then
     git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
     cd "$SRT_REPO_DIR" || exit 1
     git checkout sa-submission-q2-2026
     mkdir -p recipes/vllm/minimax-m3
     cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3" recipes/vllm/minimax-m3
-    SRTCTL_SETUP_SCRIPT="minimax-m3-vllm-fixes.sh"
+    if [[ $PRECISION == "fp8" ]]; then
+        SRTCTL_SETUP_SCRIPT="minimax-m3-vllm-fixes.sh"
+    fi
     # NVIDIA/srt-slurm#38
     git show 22d46ba9971615016d2339c9ffbc7b4597accfad --format= -- src/srtctl/core/ip_utils/get_node_ip.sh | git apply - || exit 1
-    cp \
-        "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/configs/$SRTCTL_SETUP_SCRIPT" \
-        "configs/$SRTCTL_SETUP_SCRIPT"
+    if [[ -n "$SRTCTL_SETUP_SCRIPT" ]]; then
+        cp \
+            "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/configs/$SRTCTL_SETUP_SCRIPT" \
+            "configs/$SRTCTL_SETUP_SCRIPT"
+    fi
 else
     git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
     cd "$SRT_REPO_DIR" || exit 1