Xiaomi MiMo
Xiaomi MiMo is Xiaomi's universal smart-platform and large-model initiative, developing open and deployable AI for language, multimodal understanding, voice, coding agents and physical-world/spatial intelligence. The team describes its aim as advancing human-machine collaboration and extending intelligence from the virtual world into the physical world.
At a Glance
- Software developers and engineering teams
- AI-agent and developer-tool builders
- Researchers and scientific organizations
- Enterprises needing long-context automation and multimodal workflows
- +3 more
AI Tools by Xiaomi MiMo
(2)
MiMo
Open Source Reasoning Language Model
MiMoCode
Open Source Terminal AI Agent
Discussions
No discussions yet
Be the first to start a discussion about Xiaomi MiMo
Latest News
Diagnosing and Mitigating Tool-Call Repetition in MiMo-V2.6; updated MOPD checkpoints released and API models updated
Introducing MiMo-V2.6 series; Pro and Flash open-sourced with API, AI Studio, MiMo Code, MiMo Desktop and OpenRouter availability
MiMo-V2.6-Pro demonstrated as a co-scientist for materials research and automated PFAS-capture MOF design
MiMo Code: Scaling Coding Agents to Long-Horizon Tasks
Products & Services
Open-source 7B reasoning language models trained from scratch, with pretraining, supervised fine-tuning and reinforcement-learning checkpoints; optimized for mathematics, coding and reasoning.
309B-total/15B-active sparse Mixture-of-Experts foundation model for reasoning, coding, agentic tasks and general assistance, with hybrid attention, hybrid thinking, 256K context and MTP acceleration.
Trillion-parameter-class flagship agent model with 42B active parameters, hybrid attention, MTP and 1M-token context, designed for complex workflows, tool use, coding and OpenClaw-style agents.
Omnimodal model supporting image, video and audio understanding plus agentic action through an agent scaffold; supports long continuous audio and audio-video joint reasoning.
Market Position
MiMo positions itself as an unusually open, cost-efficient alternative to frontier closed models and competing open-weight systems. Its differentiation is the combination of open weights/code, strong reasoning and coding, very long context, native multimodality, agent/tool-use optimization and inference efficiency; official comparisons reference models such as DeepSeek, Qwen, Kimi, Claude, Gemini and GPT, while the V2.6 page claims MiMo-V2.6-Pro was the highest-scoring open-source model on the cited Artificial Analysis Intelligence Index in September 2026.
Leadership
Founders
Lei Jun
Founder, Chairman, and CEO of Xiaomi. Previously President of Kingsoft and founder of Joyo.com (sold to Amazon).
Luo Fuli
Lead of Xiaomi MiMo project. Former researcher at DeepSeek and Peking University graduate. Known for leading the technical development of the MiMo series.
Executive Team
Luo Fuli
Head of the MiMo team / Xiaomi large-model team lead
Computer-science graduate of Beijing Normal University and computational-linguistics master's graduate of Peking University; previously led VECO and AliceMind work at Alibaba DAMO Academy and joined DeepSeek in 2022, where she co-developed DeepSeek-V2, before joining Xiaomi in November 2025.
Board of Directors
Founding Story
MiMo began as Xiaomi's effort to unlock stronger reasoning in relatively small language models and make the work broadly useful through open releases. The first MiMo-7B reasoning models were open-sourced on April 30, 2025; the initiative subsequently expanded toward multimodality, agentic workflows, voice and spatial/physical intelligence.
Business Model
Revenue Model
Hosted API usage billed per million tokens, with additional access through AI Studio, MiMo Code, MiMo Desktop and third-party distribution such as OpenRouter. Xiaomi also offers token plans for predictable, high-volume API usage; open-source weights and code are available for self-hosting.
Pricing Tiers
Listed on the official MiMo-V2.6 product page; pricing is in USD per million tokens.
Listed on the official MiMo-V2.6 product page; pricing is in USD per million tokens.
UltraSpeed mode for the Pro model; pricing is in USD per million tokens.
MiMo Code documentation says MiMo Auto was available free for a limited time; other access methods include Xiaomi MiMo platform login or a custom model.
Target Markets
- Software developers and engineering teams
- AI-agent and developer-tool builders
- Researchers and scientific organizations
- Enterprises needing long-context automation and multimodal workflows
- Speech and voice application developers
- Open-source AI researchers and self-hosting users
- Coding assistants and autonomous software engineering
- Long-horizon agents with hundreds or thousands of tool calls
- General-purpose chat and reasoning
- Multimodal image, video and audio analysis
- Speech recognition and expressive speech synthesis
- Scientific research and materials discovery, including literature review, molecular design and computational dry-lab experiments
- Nous Research
- OpenCode
- Ecosystem Partners in MiMo Orbit