# MiniMax M3.1 Flash weights go public and run above 1K tok/s with SGLang

> Original title: Space Bunny is here! 🐰 MiniMax M3.1 Flash hits >1K tok/s, bs1 decode with SGLang, production-ready from Day 0. This integration is a joi...

- Source: SGLang
- Published: 2026-10-23T01:09:03.000Z
- Bad Influence: https://badinfluence.news/items/eq5sibbh5ebvcd21494471cwz
- Original: https://x.com/sgl_project/status/2113437513692127424

## Summary

MiniMax has made the M3.1 Flash weights public, a model with 428B total parameters and 23B activated per token. SGLang, co-developed with NVIDIA, AMD and radixark, reports over 1K tok/s single-request decode and about 100K total tok/s per GPU on an agentic coding trace with 96 concurrent sessions. The integration supports NVIDIA Blackwell and Hopper GPUs plus AMD Instinct MI350X and MI355X, along with DSpark and DFlash2 speculative decoding.

## Why it matters

The post names the SGLang serving setup, throughput figures and hardware support, which show how the open MiniMax model runs in production.

## Post

Space Bunny is here! 🐰 MiniMax M3.1 Flash hits >1K tok/s, bs1 decode with SGLang, production-ready from Day 0. This integration is a joint effort by @MiniMax_AI, @NVIDIAAI, @AMD, and @radixark, and you can also run RL with Miles!

-Broad hardware support: NVIDIA Blackwell and Hopper (GB300/B300/B200/H200), plus AMD Instinct MI350X and MI355X.
-Agentic serving at scale: ~100K total tok/s per GPU on the AgentX coding-agent trace with 96 concurrent sessions.
-Speculative decoding: Supports MiniMax's DSpark draft and a DFlash2 draft trained by @radixark; supports HiCache and multimodal input.
-Ultra-low-latency inference: MegaKernels deliver 2.15× faster single-request decode.

[Video](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Famplify_video_thumb%2F2113437394729041920%2Fimg%2F8QvtGL5DvZBMNIRL.jpg&mode=thumb&exp=1793404800&sig=675a2563189f60f3)

## Quoted post @MiniMax_AI

> MiniMax M3.1 Flash weights are now public. 428B parameters total, 23B activated per token.
> 
> Agentic performance: A generational leap in agentic tasks, coding, and cybersecurity — 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82.2 on CyberGym.
> Efficient by design: MiniMax Sparse Attention and DSpark speculative decoding improve inference efficiency.
> 
> Hugging Face:
> https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash
> ModelScope:
> https://modelscope.cn/models/MiniMax/MiniMax-M3.1-Flash
> 
> API: http://platform.minimax.io
> M Plan: https://minimax.io/m-planMiniMax
> MCode for desktop: https://agent.minimax.io/download

https://x.com/MiniMax_AI/status/2113436864904610008