Free MiniMax M3 via the Tokenator API

What the model is for

This model is available on a free plan with pre-defined usage limits. MiniMax-M3 is a multimodal baseline model from MiniMax, focused on long-term agent work, programming, and tool use. It supports text, image, and video input, text output, and a context window of up to 1 million tokens. The model is built on the MiniMax Sparse Attention (MSA) mechanism, which uses key-value block selection instead of full attention. This significantly reduces computational overhead when working with large contexts — approximately 20x compared to the previous generation for a context of 1 million tokens — while also accelerating pre-population and decoding. MiniMax-M3 was initially trained as a multimodal model on interleaved data and is further optimized for long-term multi-step interactions. This makes it particularly well-suited for complex agent-based scenarios that require sequential execution of a large number of actions, tool use, and maintaining context throughout the task.

Specification

Upstream providers

A request goes to the first available provider by priority; if it fails, Tokenator switches to the next one.

Example request

curl https://api.tokenator.top/v1/chat/completions \
  -H "Authorization: Bearer sk-your-tokenator-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "free-minimax-m3",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Compatible tools

The model can be used in any tool that accepts a custom base URL: Claude Code, Codex CLI, OpenCode, Cursor, Cline, Kilo Code, Cherry Studio, OpenClaw.

See also