vLLM/Recipes
MiMo (Xiaomi)

XiaomiMiMo/MiMo-V2.6-Pro-RL

Xiaomi's flagship omnimodal MoE reasoning model (1.02T total / 42B active) with hybrid attention, FP8-compute/mxfp4-stored weights, 1M context, and a DFlash speculative decoder

View on HuggingFace
moe1T / 42B1,048,576 ctxvLLM nightly+text
Guide

Overview

MiMo-V2.6-Pro-RL is Xiaomi's flagship MiMo-V2.6 checkpoint, trained with large-scale mixed RL (GRPO + groupwise agentic grading) across coding, general agents, visual, and cybersecurity domains. It is a sparse MoE model with 1.02T total parameters and 42B active per token: 70 layers (1 dense + 69 MoE) with 384 routed experts (top-8), hybrid attention (sliding-window 128 and global attention at a 6:1 ratio), and a 5-layer DFlash-style MTP drafter that predicts 7 tokens per pass. The model is natively omnimodal (text, image, video, audio via a 681M-param MiMo ViT and audio encoders) and supports up to 1M tokens of context.

Weights are stored as mxfp4 and computed as FP8 (block-wise e4m3, 128x128), so the checkpoint is 566 GB on disk. Loading this mixed storage format requires vLLM with the MiMo V2 mxfp4/bf16-router support (vllm-project/vllm#57784), which is newer than the latest stable release — use the pre-built image or a nightly wheel.

Prerequisites

  • Hardware: 8x H200 (TP8) or equivalent aggregate VRAM (>= 680 GB)

Pull the vLLM docker image

Stable vLLM (<= 0.29.0) cannot load the mxfp4-stored weights. Use the pre-built image published for the MiMo-V2.6 series:

docker pull vllm/vllm-openai:mimo-v26

A vLLM nightly wheel built after 2026-09-20 also works (uv pip install -U vllm --extra-index-url https://wheels.vllm.ai/nightly/cu130).

Launch command

Single-node TP8 (H200):

vllm serve XiaomiMiMo/MiMo-V2.6-Pro-RL \
  --tensor-parallel-size 8 \
  --trust-remote-code \
  --gpu-memory-utilization 0.95 \
  --max-model-len auto \
  --reasoning-parser mimo \
  --tool-call-parser mimo \
  --enable-auto-tool-choice \
  --generation-config vllm

With DFlash speculative decoding (drafter ships inside the checkpoint):

vllm serve XiaomiMiMo/MiMo-V2.6-Pro-RL \
  --tensor-parallel-size 8 \
  --trust-remote-code \
  --gpu-memory-utilization 0.95 \
  --max-model-len auto \
  --speculative-config '{"method":"dflash","model":"XiaomiMiMo/MiMo-V2.6-Pro-RL/dflash","num_speculative_tokens":7,"draft_tensor_parallel_size":8}' \
  --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY"}' \
  --reasoning-parser mimo \
  --tool-call-parser mimo \
  --enable-auto-tool-choice \
  --generation-config vllm

Client Usage

curl -X POST http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "XiaomiMiMo/MiMo-V2.6-Pro-RL",
    "messages": [{"role": "user", "content": "Hello MiMo!"}],
    "temperature": 1.0,
    "top_p": 0.95,
    "chat_template_kwargs": {"enable_thinking": true}
  }'

Recommended sampling: temperature=1.0, top_p=0.95. Set "enable_thinking": false (or omit the kwargs) to disable thinking mode.

References