MiniMax M1 80K
M1 80K is the longer-thinking M1 weight release. It shares the Text-01-derived architecture and supports tool calls through the documented function-call format.
Long reasoning uses hybrid attention
M1 combines mixture-of-experts layers with lightning attention. The release card's main benchmark summary uses the 80K version. Its deployment guidance includes vLLM and a separate function-calling guide.MiniMax M1 80K model card
Keep long input and long reasoning separate
Practical guidance. A long input budget helps retain evidence. A long reasoning budget helps explore possible solutions. They solve different problems. Check whether extra reasoning produces a better answer before giving every request the largest budget.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | MiniMax-M1-80kMiniMax-M1-80k-hf |
| Context window | 1,000,000 tokens Native M1 context capacity; the 80K name identifies the thinking-budget release. MiniMax M1 80K model card |
| Size | 456 billion total parameters and 45.9 billion active parameters.MiniMax M1 80K model card |
| Training budget | 80,000 reasoning tokens; the 40K version is a separate trained release.MiniMax M1 80K model card |
| Licence | Apache 2.0.MiniMax M1 80K model card |
Related models
- MiniMax M3
MiniMax · MiniMax M3
A native multimodal model with sparse attention that accepts text, images and video.
Launch date API · Downloadable · Provider product - 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