MiniMax M3.1 Flash is free on AI Gateway until 20 October

On 19 October 2026, AI Gateway made MiniMax M3.1 Flash free until 20 October. The model is open-weight, and each person can get its weights. It has a context window of 1M tokens, which is the text it can read at one time. MiniMax gave no test numbers. The free rate applies only when MiniMax itself serves the task.

What is new

The model is for coding and agent work, where software does tasks of many steps. MiniMax says it is better than MiniMax models before it at complex agent tasks, coding and cybersecurity. Its best results are in app development, frontend work, small bug fixes and component tests.

The proof

The source gives no test results and no price for the model. All this comes from MiniMax. The free rate applies only when MiniMax serves the task. To make sure of this, set the provider sequence in the gateway alternatives. After the offer ends, pinned tasks have the standard MiniMax rate.

How to use it

• Use the model name minimax/minimax-m3.1-flash.
• Use the AI SDK or the Chat Completions, Responses or Anthropic Messages APIs.
• Use it in Claude Code, Codex or Cursor with the Vercel CLI.

Insight: A free model can become a habit before the price shows

Vercel says AI Gateway adds no markup and no platform fee. A team that pins a free model by name gets the full MiniMax rate after 20 October. This offer seems to turn a test into a default, and the pinned name stays in your code. Set an end date to examine each pinned model.