MiniMax M3: What It Does, Costs, and How to Use It
minimax/minimax-m3MiniMax’s open-weight foundation model takes text, image, and video inputs and produces text.
1.05M tokens of context — about 786,432 words, on a scale that ends at 2M.
What it can do
- Reasoning
- Vision
- Tool use
- Open weights
- Takes text
- Takes image
- Takes video
- Gives text
Released 31 May 2026 · checked 3 Oct 2026.
About it
MiniMax M3 is a multimodal foundation model from MiniMax that takes text, images, and video as input and produces text. It supports reasoning and tool use, with a context window of 1,048,576 tokens and a maximum output of 512,000 tokens. Its listed token prices are $0.3 per million input tokens and $1.2 per million output tokens.
Review
In a line
MiniMax M3 is an open-weight model from MiniMax. It accepts text, images, and video, and produces text.
What it's good at
M3 is suited to long-horizon agentic work and coding. It supports reasoning and tool use, and can work with image and video inputs.
Price in plain terms
Input costs $0.3 per million tokens; output costs $1.2 per million tokens. A page of about 500 words costs about 0.019 cents to read. A chat turn with about 1,000 input tokens and 500 output tokens costs about 0.09 cents; 1,000 such turns cost about $0.90.
Context and limits
M3 can take up to 1,048,576 tokens in context—the material it can consider at once—and produce up to 512,000 tokens in one answer. Its output is text, even when the input includes images or video.
Ways to use it
Its weights are open, and its Hugging Face identifier is MiniMaxAI/Minimax-M3.
Verdict
Consider M3 if you need a model that can handle text, images, or video, and you value open weights, reasoning, or tool use. Its large context and output limits suit work that spans a lot of material or requires long responses. If your task needs image or video output, M3 is not a fit: it returns text.
Verdict
MiniMax M3 is worth considering for people who need text output from text, image, or video inputs, especially for long-horizon agentic work or coding. Open weights and tool use may also matter if you are choosing a model to build into an app. The listed token rates are low in the supplied examples: 1,000 typical chat turns cost about $0.90. Choose another model if you need it to generate images or video; M3's output is text.
- Accepts text, image, and video inputs.
- Supports reasoning and tool use.
- Open weights are available.
- Context window holds up to 1,048,576 tokens.
- Maximum output is 512,000 tokens.
- Produces text only, not image or video output.
- Listed token prices are per use; total cost depends on token volume.
- Pick it ifLong-horizon agentic work involving many steps or large amounts of material; Coding tasks that benefit from reasoning and tool use; Tasks that need text responses to image or video input
- Skip it ifYou need the model to generate images or video.; You need a stated price for an app or chat plan rather than token rates.
Pricing
Token rates are $0.3 per million input tokens and $1.2 per million output tokens. The examples put a 500-word page at about 0.019 cents to read, one typical chat turn at about 0.09 cents, and 1,000 such turns at about $0.90.
Questions
What is MiniMax M3?
It is MiniMax's open-weight multimodal foundation model. It takes text, image, and video inputs and returns text.
How much does MiniMax M3 cost?
Input costs $0.3 per million tokens and output costs $1.2 per million tokens. A typical chat turn in the supplied example costs about 0.09 cents.
Can MiniMax M3 understand images and video?
Yes. It accepts both image and video inputs, but its output is text.
How much can MiniMax M3 read at once?
Its context window is 1,048,576 tokens, the amount of material it can consider at once.
How long can its answer be?
M3 can produce up to 512,000 tokens in one answer.
Where can I find MiniMax M3's Hugging Face listing?
Its Hugging Face identifier is MiniMaxAI/Minimax-M3.