MiMo V2.6 Pro RL
Pro RL is the original large V2.6 reinforcement-learning checkpoint. Xiaomi trained coding, general agents, visual tasks and cybersecurity together in one mixed run.
Research release
A useful intermediate release
The card describes asynchronous GRPO and shared training batches across tasks and agent frameworks. The later MOPD checkpoint adds another training stage, including work to mitigate repeated tool calls.MiMo-V2.6-Pro-RL official model card
Keep the training stage in the experiment record
Practical guidance. Use this checkpoint when studying the effect of the later MOPD stage. Do not label its behavior as that of the current hosted API. Keep tools and task prompts fixed so that a training-stage comparison remains meaningful.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | MiMo-V2.6-Pro-RL |
| Context window | 1,000,000 tokens Published Pro RL context capacity. MiMo-V2.6-Pro-RL official model card |
| Architecture | 1.02 trillion total parameters, 42 billion active; text, image, video and audio inputs.MiMo-V2.6-Pro-RL official model card |
| Licence | MIT.MiMo-V2.6-Pro-RL official model card |
| Training design | Groupwise grading compares candidate runs within a group, so passing answers can still receive different quality rewards.MiMo-V2.6-Pro-RL official model card |
Related models
- MiMo V2.6 Pro
Xiaomi · MiMo V2.6
MiMo V2.6 Pro accepts text, images, video and audio, then returns text.
Launch date not confirmedAPI · Downloadable · Provider product - MiMo V2.6 Flash RL
Xiaomi · MiMo V2.6
A smaller MiMo V2.6 mixture-of-experts model trained for multimodal understanding and long-context agent work.
Launch date not confirmedResearch releaseDownloadable