Alex Getmanearly AI signals

MiniMax M3 MXFP8

MiniMax

ReleasedUpdated

Timeline

  1. · Weights version · evidence link
  1. · Evidence links · GitHub search · MiniMax · commits · rodrigueztaylor87/rod · @rodrigueztaylor87
    Excerpt
    本文基于GPUStack平台,使用vLLM专用镜像完成MiniMax-M3-MXFP8模型在四卡H20-3e上的部署,支持张量并行、EAGLE3投机解码及多模态推理。实测显示开启EAGLE3后推理速度从126 tokens/s提升至200 tokens/s,加速近60%。涵盖环境配置、镜像拉取、自定义vLLM版本注册、启动参数调优(含mm-encoder-attn-backend、speculative-config等)、chat_template修复及基准测试全流程。
  2. · Evidence links · GitHub search · MiniMax · pulls · vllm-project/vllm · @afriedri
  3. · Evidence links · GitHub search · MiniMax · pulls · inference-sim/blis-vllm · @sriumcp
    Excerpt
    own `ModelConfig` over the catalog `config.json` of `deepseek-v4-pro`, `minimax-m3-mxfp8` and `minimax-m3-nvfp4`, for every method name vLLM lists, none, and a name it does not list;
  4. · Evidence links · GitHub search · MiniMax · pulls · ROCm/ATOM · @itej89
Specifications

MiniMax M3 MXFP8 is a model from MiniMax with open weights, first recorded on 7 Oct 2026; its release date is not established.

Weights
Open

Questions

Are the weights of MiniMax M3 MXFP8 publicly available?

Yes. MiniMax publishes the weights.

When was MiniMax M3 MXFP8 released?

Its release date is not established. Signal Forge first recorded it on 7 Oct 2026.