MiMo V2.6 Flash (Xiaomi)

About MiMo V2.6 Flash

MiMo-V2.6-Flash is an open-source foundation model developed by Xiaomi. It uses a Mixture-of-Experts architecture with 309B total parameters and 15B activated per token, and a hybrid attention mechanism. It became available in September 2026.

The model accepts text, image, video, and audio input and produces text. Its context window is 1,050,000 tokens and a response can be up to about 131,000 tokens. It supports reasoning, tool calling, and structured outputs. With tool calling, the free-pi agent can read and edit files, run commands, and use free-pi's built-in web search and documentation lookup tools.

It scores 37.9 on the Artificial Analysis Intelligence Index, above the default DeepSeek V4 Flash 0731 (34.3).

In free-pi it is free to use with no API key and no card. It counts as about 0.4x usage, so it uses somewhat more of your daily allowance than the default model.

When to pick it

For tasks that include audio or video as well as images and text, pick MiMo V2.6 Flash. Few free-pi models take audio, and this one uses far less allowance than Gemini 4 Argon, which also accepts audio and video. It is also an open-source model.

For text-only work that should last the day, pick the default DeepSeek V4 Flash. For a multimodal model with a larger score, pick Claude Haiku 5.5.