MiMo V2.5 API

Send text, images, video and audio to one endpoint and get text back, from $0.126.

1.1M token context
310.8B total parameters
Audio, image, video
MIT license

from openai import OpenAI

client = OpenAI(
    base_url="https://api.modelsheep.com",
    api_key=API_KEY_MODELSHEEP,
)

response = client.chat.completions.create(
    model="XiaomiMiMo/MiMo-V2.5",
    messages=[
        {
            "role": "user",
            "content": "Why do sheep count people?"
        }
    ],
)

print(response.choices[0].message.content)

How to use the MiMo V2.5 API

1
Create a ModelSheep account and copy an API key from the dashboard.
2
Point the OpenAI SDK at the base url https://api.modelsheep.com and set the model to XiaomiMiMo/MiMo-V2.5.
3
Call /chat/completions with your messages and read the text answer from the first choice, and add chat_template_kwargs with enable_thinking set to true when you want reasoning, because this model has no top level flag and leaves reasoning off.

What MiMo V2.5 is

MiMo V2.5 is an omnimodal model from Xiaomi MiMo. It reads text, images, video and audio in one prompt and answers in text. The model is a sparse mixture of experts, so each token activates a small share of its 310.8B. Xiaomi describes the larger MiMo V2.5 Pro as tuned for agent and coding work, and this model as the native full-modal member of the series.

What it does well

Answers questions about an audio file, such as a meeting recording or a lecture.
Reads a screenshot or a video clip and reports what it shows in text.
Returns tool calls, so it can run an agent loop over an API.

What it is not for

It returns text only. It does not produce audio. Xiaomi ships separate TTS models for speech.
It is not a transcription endpoint. It answers about audio in prose. A caller who needs a verbatim transcript with timestamps should use an ASR model.

Calling it

Call POST /chat/completions with the model string XiaomiMiMo/MiMo-V2.5. Audio goes in an input_audio content part, beside the text and image_url parts of the same message. Reasoning stays off unless you ask for it, and there is no top level flag, so send chat_template_kwargs with enable_thinking set to true.

XiaomiMiMo/MiMo-V2.5 /chat/completions

SDKs and languages

MiMo V2.5 answers on an HTTP endpoint, so any language that can post JSON reaches it. These are the ones the box above writes out.

OpenAI SDK: PythonTypeScriptGoJavaC#PHPRust
Anthropic SDK: PythonTypeScriptGoJavaC#PHPRust
HTTP: cURLPythonTypeScriptGoRustJavaC#PHP

Specifications

Author	Xiaomi MiMo
Category	audio
Accepts	text, audio, image, video
Returns	text
Total parameters	310.8B
Active parameters	dense, all of them
Score	22.3 on Artificial Analysis intelligence index
Languages	English, ChineseRead off the model card metadata. Xiaomi writes no sentence about language coverage.
Precision served	FP8
Context	1.1M tokens
Vector	n/a
License	MIT

Price

Input

$0.126

per million tokens

Output

$0.252

per million tokens

Against OpenRouter

10%

below the cheapest of 6 providers

The cheapest OpenRouter price for this model was $0.140 in and $0.280 out, with a median of $0.140 and $0.280 across the comparable providers. Read on 2026-09-17.

License

Xiaomi released the weights under the MIT license. It permits commercial use, fine-tuning and further training with no extra authorisation, so there are no vendor terms to read before you ship.

Parameters: 24.3B
Context: 33K
Input per 1M: $0.090
Output per 1M: $0.270
License: Apache 2.0