Xiaomi: MiMo-V2.6-Flash

MiMo-V2.6-Flash is an open-source foundation model from Xiaomi, listed on api14 under the model id xiaomi/mimo-v2.6-flash. It belongs to the text category and is built on a Mixture-of-Experts architecture with 309B total parameters and 15B activated per token, using a hybrid attention mechanism. Xiaomi positions it as a foundation model rather than a narrow task-specific system, so it is intended to serve as a general base for a wide range of language work.

The model accepts text, image, video and audio as input, and produces text as output. Its context window is 1,050,000 tokens, and the maximum output length is 131,072 tokens. That combination of a very large context and multimodal input means a single request can carry long documents alongside images, video or audio, and still return a substantial text response. Supported request parameters include temperature, top_p, top_k, min_p, frequency_penalty, presence_penalty, repetition_penalty, max_tokens, stop, seed, logit_bias, logprobs, top_logprobs, response_format, structured_outputs, tools, tool_choice, reasoning and include_reasoning.

Those capabilities point to a few practical patterns. Long-context work such as summarising or querying large document sets, codebases or transcripts fits naturally, since the window is large enough to hold material that would otherwise need chunking. Because images, video and audio can be supplied as input, the model can be used for tasks like describing media, extracting information from screenshots or scanned pages, or answering questions about recorded material, with the answer returned as text. Tool calling and structured outputs make it suitable for pipelines where the model must call functions or emit JSON that downstream code can parse, and the reasoning and include_reasoning parameters give a way to request or inspect intermediate reasoning where that is useful.

Calling it through api14 works the same way as any other OpenAI-compatible model. Send a request to the chat completions endpoint with the model field set to xiaomi/mimo-v2.6-flash, and pass messages in the usual OpenAI format, including multimodal content parts for image, video or audio input. The parameters listed above can be set in the request body; sampling controls such as temperature, top_p, top_k and min_p shape generation, max_tokens caps the output, and tools, tool_choice, response_format and structured_outputs cover function calling and structured responses. Streaming and standard OpenAI client libraries should work as expected against the compatible interface.

One thing to keep in mind is that output is text only, so the model cannot generate images, video or audio in return. If a task requires producing media rather than describing or reasoning about it, a different model would be needed. The large context and the 131,072 token output ceiling also mean requests can be heavy, so for short, simple prompts a smaller model may be a more proportionate choice.