Xiaomi MiMo API

The Xiaomi MiMo API is Xiaomi's developer platform for its MiMo models. The current lineup is MiMo-V2.6-Pro, a 1-trillion-parameter mixture-of-experts flagship, the smaller and cheaper MiMo-V2.6-Flash, and an UltraSpeed version of Pro that Xiaomi says outputs up to 20 times faster. All take text, images, audio and video, have a 1M-token context window with up to 128K tokens of output, and support thinking mode, function calls, structured output and a built-in web search tool. Separate models cover speech recognition and text-to-speech.

There's no MiMo-specific SDK to learn. The API speaks the OpenAI Chat Completions format at https://api.xiaomimimo.com/v1, the Anthropic Messages format at https://api.xiaomimimo.com/anthropic, and a Responses API, so you use the official OpenAI or Anthropic SDKs, or agent tools such as Claude Code and Codex, by swapping the base URL and key. You sign in with a Xiaomi account and usage is deducted from a prepaid balance, billed in US dollars outside mainland China and in yuan inside it.

Useful for a small team: prices are low, from $0.14 / $0.28 per million input/output tokens on Flash; prompt caching is automatic, with cache hits at about 1% of the input price and cache writes currently free; a Batch API halves the price of Pro and Flash for work that can wait; and the Pro and Flash weights are released under the MIT license, so the same models can be hosted elsewhere. Unused balance can be refunded.

Limits: there's no free quota for the language models, and each account gets 100 requests and 10M tokens per minute per model, with UltraSpeed limits by arrangement. Older model names are retired on short notice: MiMo-V2.5 and V2.5-Pro stop working on October 21, 2026. The platform is run by Beijing Xiaomi Mobile Software Co., Ltd. under Chinese law; for mainland users data is stored in mainland China, and overseas users fall under separate terms. Xiaomi's privacy policy says it won't use the text you send for model training without your prior consent.