MiniMax Launches M3.1 Flash with 3x Speed and 70% Cost Cut
MiniMax has officially launched its M3.1 Flash model, delivering a 3x increase in inference speed and a 70% reduction in cost compared to previous versions. The model achieves leading performance in agentic workflows, software development, cybersecurity, and Computer Use capabilities. This release sets a new industry benchmark by delivering top-tier model performance at a drastically reduced cost, lowering the barrier to entry for complex enterprise automation and agentic AI. Developers and organizations can now deploy large language models for demanding multi-step workflows with significantly improved efficiency. Priced at 1 RMB per million input tokens and 4 RMB per million output tokens, M3.1 Flash scored 61.7 on OSWorld V2 and 73.7% on DeepSWE v1.1. The model is currently available via the MiniMax API, MPlan, and MiniMax Code, with model weights scheduled for public release later.
## BACKGROUND
Agentic AI refers to autonomous systems capable of planning, reasoning, and using tools or software to complete complex multi-step tasks with minimal human supervision. Computer Use allows AI models to directly interact with operating systems and graphical user interfaces like a human user.