MiniMax M3.1 Flash
MiniMax released M3.1 Flash, a Flash model for agentic work and coding that it says runs nearly three times faster and costs 70% less than its predecessor, at $0.15 input and $0.60 output per 1M tokens.
M3.1 Flash is the next Flash model after M3. MiniMax says it uses more aggressive sparsity, reaching 200+ tokens per second output with a P50 of 170+. It was tested anonymously as Space Bunny, which MiniMax says processed more than 200 trillion tokens before the launch. Artificial Analysis, as reported by The Decoder, scores it 42 on its Intelligence Index, up from 29 for M3.
- Date
- Monday 19 October 2026
- Lab
- MiniMax
- Kind
- model
- Access
- closed API
- Price
- $0.15 input, $0.60 output, $0.015 cached input per 1M tokens
Figures
| Measure | Value | Measured by |
|---|---|---|
| DeepSWE v1.1 | 73.7% | company |
| OSWorld V2 | 61.7 | company |
| BankerToolBench | 76.2 | company |
| CyberGym | 82% | company |
| Agents' Last Exam (ALE) | 29.6 | company |
| Artificial Analysis Intelligence Index | 42 M3 scored 29; cost per task about $0.22; hallucination rate 93% | Artificial Analysis |
| Output speed | 200+ TPS P50 170+ TPS | company |
| Agentic task cost vs predecessor | 70% lower | company |
Available via MiniMax API, M Plan and MiniMax Code. MiniMax's page says weights will be released later, and The Decoder reports an October 23 open-weights date.
Sources
- minimax.io/blog/minimax-m31f
- the-decoder.com/minimax-m3-1-flash-shows-chinas-ai-price-race-has-no-floor-in-sight/
- forkast.news/minimax-m3-1-flash-launched-with-200-trillion-tokens-of-proof-the-name-came-s
This record was checked against its sources on 19 October 2026. How we check