MiniMax M2.7
M2.7 targets software agents and editable office work. MiniMax describes development experiments in which an internal model version helped improve its own agent software and research workflow.
Deprecated API; check access details
Self-improvement describes a training experiment
The model card describes an internal version that revised memory, created research skills and optimized an agent framework across repeated experiments. The released model also supports agent teams, dynamic tool search and work on Word, Excel and presentation files.MiniMax M2.7 model card
Keep the claim tied to its setting
Practical guidance. The development story does not mean a deployed copy trains itself during normal use. Evaluate the released weights on a fixed workflow, and keep any application memory or automatic code edits under explicit controls.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Deprecated API; check access details |
| Launch date | |
| Access | API, Downloadable |
| Model identifiers | MiniMax-M2.7 |
| API access | Closed to new users on 19 October 2026. Existing pay-as-you-go access ends on 18 November 2026. M Plan and Token Plan requests already route to M3.1 Flash.MiniMax model availability and retirement notices |
| Weight licence | Non-commercial use is permitted. Commercial use requires prior written authorization and visible MiniMax M2.7 attribution.MiniMax M2.7 non-commercial licence |
Related models
- MiniMax M3.1 Flash
MiniMax · MiniMax M3
A multimodal coding and agent model with sparse attention, quantization-aware training and a DSpark draft head.
Launch date API · Downloadable · Provider product - MiniMax M2.5
MiniMax · MiniMax M2
A coding and office model trained with reinforcement learning across more than 200,000 real-world coding environments.
Launch date not confirmedDeprecated API; check access detailsAPI · Downloadable