ModelsJul 2026
The MiniMax M-series is live — one key, three capability tiers
Hyphen now routes the MiniMax M-series through a single OpenAI-compatible endpoint. Pick a model by the job, not the vendor — flagship reasoning down to fast-and-cheap, all at roughly a tenth of frontier rates.
- m3Flagship. Long-horizon agentic coding and tool use, multimodal input, context up to ~1M tokens.
- m2.7Balanced daily driver. ~205K context; fast and strong across coding and agents.
- m2.5Efficient. Quick and cheap — ideal for high-volume, lighter calls.
Coming / on requestMiniMax M2 · M2.1 (legacy, on request)DeepSeekGLMKimi+ more