MiniMax officially releases M3.1 Flash: as low as $0.15 per million tokens, open source on the 23rd

Beating AI News Flash: MiniMax has officially released M3.1 Flash, confirming that the previously viral anonymous model Space Bunny was its early version. Space Bunny landed on OpenRouter and OpenCode in late September, surpassing 200 trillion tokens in cumulative API calls and once topping the daily call rankings on both platforms. MiniMax subsequently launched M3.1 Flash Preview on September 27, and has now officially published full benchmark scores and API pricing.

In MiniMax's published tests, M3.1 Flash scored 62.2% on the office automation benchmark Automation-Bench, surpassing DeepSeek V4.1 Flash's 60.5% and Opus 5's 50.3%. It scored 73.7% on the DeepSWE coding test, close to Opus 5's 74.0%; and reached 91.0% on the visual reasoning benchmark BabyVision. However, on the computer operation test OSWorld, its 61.7% still lags behind Opus 5's 70.6%.

Pricing is also a key focus. The API charges $0.15 per million input tokens, $0.60 for output, and only $0.015 for cache reads. MiniMax claims that compared to M3, the new model's overall cost has dropped to about one-third, with speed increased to 3x. The model supports a million-token context, and weights are scheduled to be open-sourced on October 23.