MiniMax Group Inc. has released M3.1 Flash, a faster and cheaper update to the M3 language model it introduced in June. The Chinese AI developer says agent tasks cost 70% less to run than on its previous model, while inference is nearly three times as fast.
AI agents can make dozens of model calls to complete one job, making the cost of each call consequential. MiniMax describes Flash as part of its “ongoing exploration of the balance between efficiency and capability.” The company reports output above 200 tokens per second at everyday peak.
The model’s size has changed little from M3. A developer guide for the preview release lists 428 billion parameters, and MiniMax attributes the speed gain to more aggressive sparsity during inference. Reasoning is enabled for every request. Developers can set one of five effort levels, from Low to Max. Flash accepts text, images and video, with prompts of up to 1 million tokens.
The model scored high on benchmark tests
In MiniMax’s tests, Flash scored 61.7 on OSWorld V2, a test of computer use. The company says that is the highest score for a Chinese model. It scored 76.2 on BankerToolBench, which tests banking workflows.
MiniMax also reported an 82% score on CyberGym. Researchers at the University of California at Berkeley built the benchmark to test whether AI agents can reproduce real software vulnerabilities. MiniMax lists code auditing, penetration testing and malware analysis among the model’s potential uses.
Input tokens cost 15 cents per million
Flash costs 15 cents per million input tokens, matching the off-peak price of V4.1-Flash from Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co. Ltd. DeepSeek said over the weekend that its updated model completed 77% of tasks on the DeepSWE v1.1 software engineering test. MiniMax reported 73.7% for M3.1 Flash. Neither result has been independently checked.
MiniMax has traded in Hong Kong since raising $619 million in a January initial public offering. Its shares fell after the original V4.1-Flash launched in September, Bloomberg reported.
Read nextMiniMax Releases M3.1 Flash Reasoning Model With Lower CostsSpace Bunny was an early model build
The company also confirmed that Space Bunny, an anonymous model that appeared on OpenRouter Inc.’s model routing service in late September, was an early M3.1 Flash build. Community testers had linked it to MiniMax by examining its tokenizer. More than 200 trillion tokens passed through Space Bunny while its maker remained unnamed, according to MiniMax’s launch post.
M3.1 Flash is available through the MiniMax API, the M Plan subscription and the MiniMax Code desktop app. Output costs 60 cents per million tokens. MiniMax says it will release the Flash weights later and has a more capable Pro model coming.
