Frequently Asked Questions
Everything you need to know about Xiaomi MiMo — licensing, models, pricing, deployment, and more.
Is Xiaomi MiMo open source?
Yes. All Xiaomi MiMo models are released under the MIT License — one of the most permissive open-source licenses. Weights, code, and tooling are freely available on Hugging Face and GitHub. You can use MiMo commercially, fine-tune it, and distribute it without restrictions.
What models are available?
The full lineup: MiMo-7B (lightweight reasoning), MiMo-V2-Flash (MoE flagship, 309B total/15B active), MiMo-V2.5-Pro (1T+ agent flagship, 1M context), MiMo-V2.5-Omni (full-modal), MiMo-V2.5-TTS (speech), MiMo Code (terminal coding agent), MiMo-VL (vision-language), MiMo-Audio, and MiMo-Embodied (robotics). See full details →
How does API pricing work?
Pricing (permanent since May 2026):
- MiMo-V2.5-Pro: $1/M input + $3/M output tokens
- MiMo-V2-Flash: $0.50/M input + $1.50/M output tokens
For comparison, GPT-5.5 is $5/$30 and Claude Opus 4.5 is $15/$75. MiMo is 80–95% cheaper. Self-host the open weights and the marginal inference cost drops to zero. API guide →
What is MiMo Code?
MiMo Code is Xiaomi's terminal-native AI coding agent, released June 2026 and MIT licensed. Features include infinite context, persistent memory across sessions, multi-agent switching (ask/architect/edit/debug modes), and integration with VS Code, Cursor, Cline, and Zed. Read the full guide →
Can I deploy MiMo on device?
Absolutely. MiMo is optimized for edge deployment from the ground up:
- MiMo-7B — runs on phones, smart speakers, and car cockpits with INT4 quantization
- V2-Flash — deployable on GPU-equipped edge servers (150 tok/s)
- V2.5-Pro — best suited for cloud access via API or GPU clusters
How does MiMo compare to other models?
MiMo — especially V2-Flash — competes directly with open-weight models like Llama 4, DeepSeek, and Qwen, while undercutting GPT-5.5 and Claude Opus 4.5 on pricing. Key differentiators: MIT license (no restrictions), edge-first design (runs on-device), and the Human-Car-Home ecosystem integration. MiMo vs DeepSeek →
How do I access the MiMo API?
Sign up at platform.xiaomimimo.com. The API is fully OpenAI-compatible — just change the base_url in your existing OpenAI client library to https://api.xiaomimimo.com/v1. Full API guide →
What benchmarks does MiMo have?
- SWE-Bench verified: 73.4% (MiMo-V2-Flash)
- AIME 2024: 68.2% (MiMo-7B-RL)
- Context window: 1M tokens (MiMo-V2.5-Pro)
- Inference speed: 150 tok/s (MiMo-V2-Flash)
Where can I download the models?
All MiMo model weights and code are on Hugging Face and GitHub. MiMo Code installs via curl -fsSL https://mimo.xiaomi.com/install | bash.
Is there an official mobile app?
Xiaomi's Miclaw app provides MiMo-powered AI assistant features on Xiaomi phones (primarily in China). For international users, the main access points are the cloud API and the open-source model weights.
Who built MiMo?
MiMo is developed by Xiaomi's LLM Core Team and AI Lab, led by Luo Fuling, supported by Xiaomi's ¥40B (~$5.5B) AI investment. The model family powers the Human-Car-Home ecosystem — Xiaomi's strategy to embed AI across 500M+ devices.
Where can I find official documentation?
Xiaomi maintains official MiMo documentation at mimo.mi.com/docs (primarily in Chinese). Additional resources: HuggingFace, GitHub, mimo.xiaomi.com.