MiniMaxAI/MiniMax-M3.1-Flash
just landedmodel
MiniMax-M3.1-Flash is a native multimodal mixture-of-experts model for coding, agentic workflows, computer use, and cybersecurity. Mixture-of-experts, 220B total, 23B active (128 experts, 4 per token). Context: 1M tokens (config). License: minimax-community.
- Org
- MiniMaxAI
- Released
- 2026-10-17
- Status
- just landed
- License
- minimax-community (restricted)
- Parameters
- 220.5B total, 23.0B active
- Architecture
- MoE · 128 experts, 4/token · MiniMaxM3SparseForConditionalGeneration
- Context
- 1,048,576 tokens
- Task
- image-text-to-text · vision
- Signal
- 336 HF likes · 7355 downloads · 1104 X likes · 3 X post(s)
Runs on
mixed: real files + estimate; need = weights + FP16 KV cache for 8K tokens + 1GB. A sizing calculation, not tested.
- Q4 ~134.3 GB (estimated size): 2× DGX Spark (256GB), Mac 192GB, Mac 256GB, CPU only, 256GB RAM (slow)
- Q8 ~261.1 GB: Mac 512GB
- FP16 ~443.0 GB (estimated size): no listed device
Links
- Hugging Face
- Ollama
- MLX: mlx-community/MiniMax-M3.1-Flash-Text-3bit
- MLX: mlx-community/MiniMax-M3.1-Flash-Text-4bit
- MLX: mlx-community/MiniMax-M3.1-Flash-oQ3.5e-mtp
- MLX: mlx-community/MiniMax-M3.1-Flash-oQ3e-mtp
- MLX: mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp
- Unsloth: unsloth/MiniMax-M3.1-Flash
- GGUF: unsloth/MiniMax-M3.1-Flash-GGUF
First published here 2026-10-17 · updated 2026-10-23 · markdown · JSON