MiniMax M3.1 Flash

Agentic AI task costs drop by 70% for developers using the MiniMax API.

MiniMax Research has launched its M3.1 Flash model, an advancement in large language model capabilities, particularly for agentic tasks. The model, which has processed over 200 trillion tokens, demonstrates significant improvements in areas like complex agentic tasks, coding, computer use, and cybersecurity. It achieves performance comparable to leading models on benchmarks such as AutomationBench-AA and BankerToolBench for enterprise workflows, DeepSWE v1.1 for software engineering, OSWorld V2 and Agents’ Last Exam (ALE) for computer operations, and CyberGym for cybersecurity. M3.1 Flash offers output speeds exceeding 200 tokens per second and reduces usage costs by approximately 70% compared to its predecessor. This new model is now accessible via the MiniMax API, M Plan, and MiniMax Code, with model weights to be released later.

MiniMax's M3.1 Flash release signals a significant market push towards more cost-effective and capable AI agents. By drastically lowering usage costs and boosting inference speeds, MiniMax is making advanced AI automation accessible for a broader range of enterprise applications and developers. This move challenges competitors to balance performance with economic viability, potentially accelerating adoption of AI in everyday business workflows across sales, marketing, finance, and HR.

The model's strong performance across diverse benchmarks like complex workflows, coding, computer operation, and cybersecurity indicates a maturing AI agent technology capable of handling real-world, multi-step tasks. This enhanced ability to interact with systems and execute sequences of actions suggests a future where AI agents can autonomously manage significant portions of enterprise operations, reducing the need for extensive human oversight on routine and complex processes.