MiniMax M3.1 Flash is built for agents
A new fast model from MiniMax aimed at agent work at a lower cost.
MiniMax has released M3.1 Flash, a model aimed at agent work such as multi-step tasks, coding, operating a computer and security analysis. The company says an early version ran anonymously under the name Space Bunny and processed more than 200 trillion tokens before the launch.
For agents that call a model many times per task, cost and speed decide what is practical. MiniMax says M3.1 Flash makes online use nearly three times faster and about 70 percent cheaper to use than its predecessor, with typical output above 170 tokens per second. Benchmark results such as 73.7 percent on DeepSWE v1.1 are the company's own.
The model costs $0.15 per million input tokens and $0.60 per million output tokens. It is available through the MiniMax API, M Plan and MiniMax Code. MiniMax says the weights will be released later, and a larger Pro model is also coming.