MiniMax M3

Overview

MiniMax M3 is a multimodal foundation model from MiniMax, released on 2026-05-31. It accepts text, images and video, and produces text. Its 1,048,576-token context window gives it room to work with long inputs, while support for reasoning and tools makes it relevant to agentic workflows as well as coding tasks.

M3 is listed as an open-weight model, with the Hugging Face identifier MiniMaxAI/Minimax-M3. Its model ID is minimax/minimax-m3. For builders, those identifiers and its input and output modalities are useful starting points when assessing where it may fit in an existing stack.

Explore other MiniMax models, open-weight models, vision language models, reasoning models, longest-context models and newest models.

Key features

Multimodal input, text output

M3 accepts text, image and video inputs, but its listed output is text. That distinction matters when a workflow needs to interpret visual material and return a written result, rather than generate images or video.

Long context and agentic work

The model has a context window of 1,048,576 tokens and a maximum output of 512,000 tokens. MiniMax describes it as suited to long-horizon agentic work and coding. Tool support and reasoning are both listed, although the available details do not specify particular tools, capabilities or task-level performance.

Open weights

M3 is marked as open weight, and its listed Hugging Face ID is MiniMaxAI/Minimax-M3. Teams evaluating model access can use that status and identifier as part of their deployment and stack-fit assessment; no further license or hosting terms are specified here.

Pricing

Listed token pricing is USD $0.30 per 1 million input tokens and USD $1.20 per 1 million output tokens. Input and output are priced separately, so the total depends on both how much content a request sends and how much text the model returns.

Platforms

The listed model ID is minimax/minimax-m3, and the model is associated with MiniMax. Its official listing is at openrouter.ai/minimax/minimax-m3. The available information does not specify additional platform integrations or deployment options.

Who it's for

M3 is worth evaluating for teams building workflows that need to process long inputs, use images or video alongside text, or connect model reasoning with tools. Its stated fit for coding and long-horizon agentic work also makes it relevant to developers considering those use cases. Teams should account for the text-only output format and verify that the model's unspecified tool behavior meets their requirements.

The unusually large context and output limits may matter to applications that need to keep extensive material in a single interaction. They are capacity limits, not a guarantee of quality or of practical performance for any particular workload.

Pros and cons

Pros

• Accepts text, images and video inputs.
• Has a 1,048,576-token context window and a 512,000-token maximum output.
• Listed as open weight, with a Hugging Face identifier.
• Reasoning and tool support are listed.

Cons

• Produces text only, not image or video outputs.
• Specific tool capabilities, licensing terms and deployment choices are not detailed.
• Token prices are not the only factor in cost: actual spend depends on usage.

Alternatives

For other options in the MiniMax family, consider MiniMax M2.7, MiniMax M2.5, MiniMax M2-her, MiniMax M2.1, MiniMax M2, MiniMax M1 and MiniMax-01.

Verdict

MiniMax M3 combines multimodal input, text output, reasoning, tools and a very large context window, with open-weight status and clearly listed per-token prices. That makes it a candidate for teams exploring long-context coding or agentic workflows that also need to interpret images or video. The key limitations in the available details are the text-only output and the lack of specifics about tools, licensing and deployment. Assess those requirements alongside token usage before choosing it for a production stack.