MiniMax releases M3.1 Flash and says agent tasks cost 70% less than on M3, at $0.15 per million input tokens

MiniMax, the Chinese AI developer, released M3.1 Flash on 19 October, a faster and cheaper update to the M3 model it launched in June. Its launch post says: “For agentic tasks, the model's overall usage cost is 70% lower than its predecessor.” It also says online inference is nearly three times faster, with output above 200 tokens per second in everyday use.

The flat API price is $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. The model is available through the MiniMax API, its M Plan subscription and the MiniMax Code app. The post says: “Model weights will be released later.” It gives no date.

On MiniMax's own tests, M3.1 Flash scores 61.7 on OSWorld V2, a test of agents using a computer, and 73.7% on DeepSWE v1.1, a coding test. MiniMax says its OSWorld score is the best among Chinese models. None of these scores has been checked independently.

MiniMax also confirmed that an anonymous model on OpenRouter, called Space Bunny, was an early build of M3.1 Flash. SiliconANGLE reported the model has 428 billion parameters, citing MiniMax's developer guide; the launch post itself gives no size.

• Claimed Usage cost for agentic tasks is 70% lower than M3, and online inference is nearly three times faster, MiniMax says. MiniMax
• Confirmed The API price is $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens; the weights “will be released later”. MiniMax
• Reported The model has 428 billion parameters, SiliconANGLE reported, citing MiniMax's developer guide. SiliconANGLE