MiniMax M3.1 Flash cuts token prices, but still needs a careful migration

A cheaper model is useful only if the finished job stays cheaper. MiniMax’s October 19 M3.1 Flash rollout gives developers a new token-price starting point: $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. The official pricing table confirms those flat rates.

An earlier preview already existed. This is coverage of the dated rollout and its current terms, rather than a claim that nobody could use a preview before October 19. MiniMax says access is through its API, M Plan and MiniMax Code, with model weights to follow. Its speed and task-cost comparisons remain vendor claims, not results measured by CreatorStack. Launch announcement.

Check the adapter before the invoice

The model table lists a one-million-token context window. In the Anthropic-compatible interface, the exact identifier is MiniMax-M3.1-Flash. Thinking is always on; disabling it produces an error. Effort runs from low through max, defaulting to max, and thinking tokens count toward the output limit. Multi-turn tool workflows must retain the required assistant content blocks. Compatibility and thinking controls.

That matters if your existing integration treats every assistant reply as plain text. A proposed migration check should include a tool call, its result and a follow-up question. Inspect the final user-facing answer and the response accounting. A successful first sentence does not prove the next tool turn will work. We have not executed this check.

Price a completed task

As illustrative arithmetic, two million uncached input tokens plus half a million billed output tokens cost $0.60 at the listed rates: $0.30 + $0.30. That is a token subtotal, not a measured project bill. Retries, actual reasoning usage and other services can change the total; an M Plan subscription also has its own entitlement rather than that API invoice.

Before switching a creator-site workflow, keep one reproducible edit and its acceptance checks fixed. Compare the accepted change, total usage and time spent correcting it. Our AI output evaluation guide explains how to judge that result; the tool-spending guide helps put a boundary around the trial.