MiniMax Text-01
Text-01 is MiniMax's January 2025 language-model release. It combines lightning attention, softmax attention and a mixture-of-experts network to handle long inputs.
Attention and parallelism were developed together
The card names LASP+, variable-length ring attention and expert tensor parallelism among its training methods. It also documents structured function calling.MiniMax Text-01 model card
The model licence has its own conditions
Redistribution requires the agreement, attribution and notices for modified files. The model agreement includes a prohibited-use policy; the code's MIT label does not replace these weight terms.MiniMax Text-01 weight licence
Validate evidence across the full input
Practical guidance. Large context capacity does not prove that every detail remains equally accessible. Put important facts at several positions in a document set, then check whether the model retrieves and cites them correctly.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | Downloadable |
| Model identifiers | MiniMax-Text-01MiniMax-Text-01-hf |
| Context window | 4,000,000 tokens Inference context described by the model card, extended from one-million-token training. MiniMax Text-01 model card |
| Size | 456 billion parameters in total; 45.9 billion active per token.MiniMax Text-01 model card |
| Context distinction | Trained at one million tokens; the card describes inference extension to four million tokens.MiniMax Text-01 model card |
| Licences | Model weights use the custom MiniMax Model License. The separate code licence is MIT.MiniMax Text-01 model card |
Related models
- 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 - MiniMax VL-01
MiniMax · MiniMax-01
A Text-01 vision-language adaptation with a 303M vision transformer and dynamic image patches.
Launch date not confirmedDownloadable