← Catalog
MiniMax M3.1 Flash
AvailableMiniMaxOpen weights
MiniMax M3.1 Flash is MiniMax's fast MoE model in the M3.1 generation. It first appeared 2026-09-27 as MiniMax-M3.1-Flash-Preview (Token Plan / MiniMax Code only), launched publicly on 2026-10-19, and its weights were published 2026-10-23 on Hugging Face (MiniMaxAI/MiniMax-M3.1-Flash, MXFP8, ~252 GB) under the MiniMax community license. Mixture-of-Experts, ~428B total / ~23B active, 1M-token context, text + image + video input, always-on thinking with five effort levels. API $0.15 / $0.60 per 1M input/output ($0.015 cached input).
Specifications
- License
- Open weights · MiniMax Community License
- Weights
- Downloadable
- Architecture
- Mixture-of-Experts
- Parameters
- 428B · 23B active
- Context window
- 1.0M tokens
- Max output
- —
- Knowledge cutoff
- —
- Price (in / out, $/M)
- $0.15 / $0.6
- Modalities
- TextVisionVideoCode
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.