MiMo V2.6 Flash RL
Flash RL is the smaller original V2.6 reinforcement-learning release. It combines four input modalities with a mixture-of-experts network and long-context agent training.
Research release
Flash retains the multimodal front end
Its card lists a 681-million-parameter vision encoder, an audio tokenizer plus patch encoder, and a five-layer speculative decoder. These support the same four input types described for Pro.MiMo-V2.6-Flash-RL official model card
Use it to study the update, not to infer current API behavior
Practical guidance. The earlier checkpoint can help expose which tool behaviors changed after distillation. Record both redundant actions and task success. A final correct answer can still come from an inefficient or unstable sequence.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | MiMo-V2.6-Flash-RL |
| Context window | 1,000,000 tokens Published Flash RL context capacity. MiMo-V2.6-Flash-RL official model card |
| Size | 309 billion total parameters, 15 billion active per token.MiMo-V2.6-Flash-RL official model card |
| Licence | MIT.MiMo-V2.6-Flash-RL official model card |
| Training signals | The groupwise reward method scores solution quality as well as whether a task passes.MiMo-V2.6-Flash-RL official model card |
Related models
- MiMo V2.6 Flash
Xiaomi · MiMo V2.6
A multimodal agent model that accepts text, images, video and audio and returns text.
Launch date not confirmedAPI · Downloadable · Provider product - MiMo V2.6 Pro RL
Xiaomi · MiMo V2.6
A large multimodal mixture-of-experts model trained jointly on coding, agent, visual and cybersecurity tasks.
Launch date not confirmedResearch releaseDownloadable