Xiaomi MiMo V2-Pro AI model release
Xiaomi, the third-largest smartphone manufacturer globally, has a market cap of around $137 billion and shipped roughly 170 million phones in 2025. On March 18, Xiaomi released the Xiaomi MiMo V2-Pro AI model alongside MiMo-V2-Omni and a text-to-speech model. Xiaomi described these releases as its first full-stack model family built for the Agent era, and they accompany milestones including the SU7 Ultra Nürburgring record set last year, the December 2025 release of MiMo-V2-Flash, and Hunter Alpha appearing on OpenRouter on March 11 as an early internal test build of MiMo-V2-Pro.
MiMo-V2-Pro comprises over one trillion total parameters and uses a mixture-of-experts design that activates about 42 billion parameters per request. The model’s context window can handle up to one million tokens, supporting long-range inputs. It implements a hybrid attention mechanism at a 7:1 ratio. These architectural choices are part of the model’s reported technical specifications.
MiMo-V2-Pro includes a built-in multi-token prediction layer that speeds up generation. The model is currently closed source, although Xiaomi left open the possibility of releasing it in the future. On the Artificial Analysis Intelligence Index, MiMo-V2-Pro ranks eighth worldwide and second among Chinese models, behind GLM-5. On SWE-bench Verified, MiMo-V2-Pro scored 78%, compared with Claude Opus 4.6 at 80.8% and Claude Sonnet 4.6 at 79.6%.
The technical specifications above summarize the reported capabilities and verified benchmark results for MiMo-V2-Pro. The release materials also note the model’s current closed-source status. These documented metrics and availability details form the basis for assessments of the model’s technical profile.
MiMo-V2-Flash was released in December 2025 as a 309B mixture-of-experts model. On March 18, Xiaomi released MiMo-V2-Pro alongside MiMo-V2-Omni and a text-to-speech model. The December release and the March 18 launch form part of Xiaomi’s recent MiMo family rollout. These product milestones provide a chronological context for the MiMo-V2-Pro announcement.
Hunter Alpha appeared on OpenRouter on March 11 and was described as a 1-trillion-parameter model before later being confirmed as an early internal test build of MiMo-V2-Pro. The Hunter Alpha reveal preceded Xiaomi stock jumping 5.8%. The sequence of an externally visible model appearance followed by confirmation of its connection to MiMo-V2-Pro is documented in reporting. These events occurred in the lead-up to the public release of MiMo-V2-Pro.
The items above constitute the documented competitive and market context surrounding MiMo-V2-Pro. They are limited to the facts presented in the release materials and reporting.
Xiaomi has expanded its MiMo model family with the public release of MiMo-V2-Pro alongside MiMo-V2-Omni and a text-to-speech model, following the earlier launch of the MiMo-V2-Flash model. MiMo-V2-Pro incorporates a mixture-of-experts architecture with a large total parameter footprint and a subset active per request, a context window designed for very long inputs, a hybrid attention mechanism, and a built-in multi-token prediction layer, and the model is currently closed source though Xiaomi left open the possibility of a future release.
