MiniMax M2.1
M2.1 is an available weight release focused on multilingual software work, tool use and longer plans. Its card describes improvements over M2 in code review, performance optimization and instruction following.
Framework results are part of the evidence
MiniMax evaluated M2.1 with several coding-agent frameworks. The model card also describes an internal code-review evaluation that penalizes false findings, not only missed defects.MiniMax M2.1 model card
Check behavior in your own development tools
Practical guidance. A useful review assistant must find defects without flooding the user with incorrect warnings. Compare both missed issues and false findings. Keep the agent framework fixed when comparing M2.1 with a newer model.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | Downloadable |
| Model identifiers | MiniMax-M2.1 |
| API status | The current lifecycle page lists the original M2.1 API as retired; this guide covers the downloadable weights.MiniMax model availability and retirement notices |
| Licence | Modified MIT: commercial products or services must display MiniMax M2.1 in their user interface.MiniMax M2.1 modified MIT licence |
Related models
- 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 - MiniMax M2
MiniMax · MiniMax M2
A coding and tool-use model with 230B total parameters and 10B active parameters per token.
Launch date Downloadable