MiMo-V2.5 is a next-generation native multimodal AI model developed by Xiaomi MiMo, designed for AI Agents, multimodal reasoning, and complex real-world applications.
The model supports unified understanding across text, images, videos, and audio, enabling advanced interactions across multiple data types. Built with a sparse Mixture-of-Experts (MoE) architecture, MiMo-V2.5 features approximately 310B total parameters with around 15B active parameters, delivering strong performance with improved computational efficiency.
With support for up to 1M-token context windows, MiMo-V2.5 can process long documents, multimodal inputs, and complex workflows. It also includes advanced reasoning, tool calling, web search, and structured output capabilities for building intelligent AI applications.
As part of Xiaomi’s open-source AI model ecosystem, MiMo-V2.5 is released under the MIT License, allowing developers and organizations to customize, deploy, and build commercial AI solutions.
MiMo-V2.5-Pro
76 MiMo-V2.5-Pro is a flagship MoE model specialized in long-term agent tasks and complex coding. It stably executes over 1,000 tool calls, reduces token consumption by 40–60% vs. competitors, and is open-sourced under MIT license.
Xiaomi
Comments (0)