Xiaomi MiMo-V2.6 Series Models Officially Launch on B.AI, Flagship All-Modal and Efficient Versions Now Open for Both Platforms

B.AI has announced the full launch of Xiaomi's open-source, native multimodal reasoning large models in the MiMo-V2.6 series, with simultaneous access via both API and Web Chat. The entire series features a 1M token ultra-long context window and comprehensive understanding capabilities across text, images, video, and audio. The MiMo-V2.6-Pro utilizes a sparse MoE architecture with 1.02 trillion total parameters and 42 billion active parameters, specifically engineered for repository-level software engineering, long-horizon agents, and complex multimodal reasoning. The MiMo-V2.6-Flash, on the other hand, is built on an architecture with 309 billion total parameters and 15 billion active parameters, prioritizing high throughput and rapid response for high-frequency office automation and large-scale agent invocation scenarios. Starting today, developers and users can directly invoke both models via the B.AI platform. Visit chat.b.ai/chat to experience them now.