MiniMax launches M3.1 Flash model with faster inference, lower costs, and later planned weight release

MiniMax says M3.1 Flash is now available through its API, M Plan, and MiniMax Code, with model weights to be released later. The company claims the model cuts agentic-task usage costs by 70% versus its predecessor, reaches 200+ tokens per second in everyday use, and scores 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, and 82% on CyberGym.

For teams building agentic workflows, the reported $0.15 input and $0.60 output pricing per 1M tokens could make high-step automation and coding tasks cheaper to trial. The trade-off is deployment control: access is available through MiniMax channels now, but weights are only promised later, so self-hosting or deeper model governance remains an unanswered question.