Minimax M3 API

A multimodal mixture of experts model for agentic coding, long document work and video input, from $0.252 per million input tokens.

Get API key Read the docs

1M context window
427B total parameters
23B active parameters
Text, image, video

from openai import OpenAI

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

response = client.chat.completions.create(
    model="MiniMaxAI/Minimax-M3",
    messages=[
        {
            "role": "user",
            "content": "Why do sheep count people?"
        }
    ],
)

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

How to use the Minimax M3 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 MiniMaxAI/Minimax-M3.
3
Call /chat/completions with your messages and send thinking: {"type": "disabled"} to stop the reasoning that runs by default, then read the reply from the first choice.

What Minimax M3 is

Minimax M3 is an open weight mixture of experts model from MiniMax, released in June 2026. MiniMax trained it on text, image and video together from the first step, so it reads all three in one request. A sparse attention design called MiniMax Sparse Attention carries the 1M token window at a fraction of the compute full attention needs. Against MiniMax M2 it adds image and video input, the long window, and higher reported scores on coding and agent benchmarks.

What it does well

Works a repository issue across many tool calls and returns a patch.
Answers questions over a long codebase, a long document or a long video in one request.
Reads charts and formulas out of a paper and writes them back as text.

What it is not for

Short, low latency calls. The model reasons before it answers by default on the chat completions endpoint, and those reasoning tokens bill as output tokens.
Shipping under a standard open license. MiniMax Community sets conditions on commercial use. Read them first.

Calling it

Post to /chat/completions with MiniMaxAI/Minimax-M3 as the model string. Images and video go in the message content as image_url parts, the shape the OpenAI SDK already sends. Reasoning is the one thing callers get wrong. It runs by default, so send thinking: {"type": "disabled"} to turn it off, then check the response for reasoning content, because some serving stacks drop the field and keep reasoning on.

MiniMaxAI/Minimax-M3 /chat/completions

SDKs and languages

Minimax M3 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	MiniMax
Category	language
Accepts	text, image, video
Returns	text
Total parameters	427B
Active parameters	23B
Score	29.6 on Artificial Analysis intelligence index
Languages	not published
Precision served	FP8
Context	1M tokens
Vector	n/a
License	MiniMax Community

Price

Input

$0.252

per million tokens

Output

$0.990

per million tokens

Against OpenRouter

10%

below the cheapest of 11 providers

The cheapest OpenRouter price for this model was $0.280 in and $1.100 out, with a median of $0.300 and $1.200 across the comparable providers. Read on 2026-09-17.

License

MiniMax publishes M3 under the MiniMax Community license, not a standard open license such as Apache 2.0 or MIT. It permits commercial use, but a company above the stated yearly revenue threshold needs written permission from MiniMax first, and a commercial product must display "Built with MiniMax M3". The license also bars military use, use that harms minors, and content that promotes discrimination. Read the license file before you ship.

Parameters: 1.6T, 37B
Context: 1M
Input per 1M: $0.939
Output per 1M: $1.879
License: MIT