# MiniMax releases M3.1-Flash, a native multimodal MoE model on Hugging Face

> Original title: MiniMaxAI/MiniMax-M3.1-Flash

- Source: MiniMax · new models on Hugging Face
- Published: 2026-10-17T06:37:50.000Z
- Bad Influence: https://badinfluence.news/items/f08iy2x24vnndbt6tfm1kvll5
- Original: https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash

## Summary

MiniMax has released M3.1-Flash on Hugging Face, a native multimodal mixture-of-experts model with about 428B total and 23B activated parameters per token. It accepts text, image, and video inputs, supports a context window of up to 1,000,000 tokens, and uses sparse attention, quantization-aware training, and DSpark speculative decoding. The model card reports results such as 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld 2.0, and 82.2 on CyberGym, and recommends SGLang, vLLM, or TokenSpeed for serving.

## Why it matters

The model card lists architecture, context length, and benchmark results against several named rivals, letting readers compare it with current open and closed models.