MiniMax M2
M2 is a text model for coding and tool-based agents. Its mixture-of-experts design has 230 billion total parameters and activates 10 billion per token.
The model was designed around agent work
The card describes multi-file editing and repeated code, execution and repair steps. Its coding evaluation used a specific agent framework, a 128K evaluation context and a limit of 100 steps. Those settings describe the evaluation, not a general context limit.MiniMax M2 model card
Active size does not equal stored size
Practical guidance. Do not plan memory from the 10-billion active-parameter figure alone. Local serving must also account for the stored experts, runtime data and request length. Compare total resource use against the number of useful tasks completed.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | Downloadable |
| Model identifiers | MiniMax-M2 |
| Original API | Retired in the current API lifecycle table. Published model weights remain available.MiniMax model availability and retirement notices |
| Weight licence | Modified MIT. Commercial attribution is required above 100 million monthly active users or US$30 million in annual recurring revenue.MiniMax M2 modified MIT licence |
Related models
- MiniMax M2.1
MiniMax · MiniMax M2
A downloadable model for multilingual software work, code review, tool use and longer plans.
Launch date Downloadable - MiniMax M1 40K
MiniMax · MiniMax M1
A Text-01-derived reasoning model trained with a 40,000-token thinking budget and lightning attention.
Launch date not confirmedDownloadable