---
type: "Article"
title: "MiniMax-M3.1-Flash NVFP4/MXFP8 with vLLM on 4× DGX Spark"
description: "MiniMax-M3.1-Flash NVFP4/MXFP8 with vLLM on 4× DGX Spark: recipe by tonyd2wild with speed, condition and source for every number, weights and license."
resource: "https://www.contextstudios.ai/local-ai/tonyd2wild--minimax-m3.1-flash-tp4-dspark-300k-4x-dgx-spark"
language: "en"
generated:
  by: "process:contextstudios-md/1"
  at: "2026-11-11T03:30:17.848Z"
status: "stable"
---

# MiniMax-M3.1-Flash NVFP4/MXFP8 with vLLM on 4× DGX Spark

MiniMax-M3.1-Flash NVFP4/MXFP8 with vLLM on 4× DGX Spark: 22.2 tok/s according to github.com (dataset as of Nov 3, 2026).

- Creator: [tonyd2wild](https://www.contextstudios.ai/local-ai/creators/tonyd2wild.md)
- Engine: vLLM 0.1.dev23341+g8744dd2fb (vllm/vllm-openai:minimax-m31-flash-arm64-cu134)
- Quantization: NVFP4 experts (Marlin W4A16 on GB10) + MXFP8 attention/dense, FP8 KV (official checkpoint)
- Model family: MiniMax-M3.1-Flash
- Hardware: dgx-spark x4
- Context: 300000

## All numbers

- 22.2 tok/s (Everyday; Everyday, 1 request, short prose prompt, with DSpark (per recipe)) — [source](https://github.com/tonyd2wild/MiniMax-M3.1-Flash-TP4-DGX-Sparks#speed-c1-to-c6)
- decode_tps: 22.2 (1 request) — [source](https://github.com/tonyd2wild/MiniMax-M3.1-Flash-TP4-DGX-Sparks#speed-c1-to-c6)
- decode_tps: 33.2 (1 request) — [source](https://github.com/tonyd2wild/MiniMax-M3.1-Flash-TP4-DGX-Sparks#speed-c1-to-c6)
- prefill_tps: 1474 (1 request) — [source](https://github.com/tonyd2wild/MiniMax-M3.1-Flash-TP4-DGX-Sparks#speed-c1-to-c6)
- ttft_ms: 520 (1 request) — [source](https://github.com/tonyd2wild/MiniMax-M3.1-Flash-TP4-DGX-Sparks#speed-c1-to-c6)

## Weights

- [MiniMaxAI/MiniMax-M3.1-Flash](https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash) (main, other)

[https://github.com/tonyd2wild/MiniMax-M3.1-Flash-TP4-DGX-Sparks](https://github.com/tonyd2wild/MiniMax-M3.1-Flash-TP4-DGX-Sparks)

## Related

- [Local AI on your hardware.](https://www.contextstudios.ai/local-ai.md)