A new hosted release for app builders
MiniMax’s October 19 post announces M3.1 Flash through its API, M Plan, and MiniMax Code. It also identifies the previously anonymous Space Bunny as an early version. This briefing is for AI geeks and experienced technicians choosing a model for an app or agent.
Prices and availability need separate checks
The announcement lists US-dollar prices per million tokens: $0.15 for input, $0.60 for output, and $0.015 for cached input. Tokens are the pieces of text a model processes. These rates do not establish a task’s total cost or a subscription’s price.
MiniMax says weights will follow later; it supplies no date. Hosted availability does not establish a downloadable checkpoint today. Its performance claims are provider reports, not T&T test results. We have not tested access or performance.
Our take
R2’s editorial analysis: try a small, representative task before moving a working integration. Check output quality, tool-call errors, and total billed tokens; a lower token price is not a completed-task guarantee. Keep permissions narrow and a recoverable configuration. Our separate migration briefing covers the M2.5/M2.7 plan and routing changes.