MiMo V2.5 Pro Base
V2.5 Pro Base exposes the large pretrained text model for further research. Its download table gives a 256K context, below the one-million-token post-trained release.
Research release
Read the checkpoint row rather than the family overview
The shared card discusses the final Pro family, but its base-model download row states 256K. The base and post-trained checkpoints therefore have separate guides and context records.MiMo-V2.5-Pro-Base official model card
A base checkpoint needs its own behavior checks
Practical guidance. Use this release for continued training or controlled comparisons. Do not assume that a base model follows instructions or calls tools like the final agent model. Evaluate those behaviors after the intended adaptation.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | MiMo-V2.5-Pro-Base |
| Context window | 262,144 tokens Base download table lists 256K context. MiMo-V2.5-Pro-Base official model card |
| Size and format | 1.02 trillion total parameters, 42 billion active; mixed FP8 E4M3 release.MiMo-V2.5-Pro-Base official model card |
| Licence | MIT.MiMo-V2.5-Pro-Base official model card |
Related models
- MiMo V2.5 Pro
Xiaomi ยท MiMo V2.5
V2.5 Pro is a text-only mixture-of-experts release for software engineering and long agent runs.
Launch date not confirmedDownloadable