MiniMax M3.1 Flash (www.minimax.io)

MiniMax has launched its latest AI model, M3.1 Flash, marking a significant evolution in capabilities for agentic tasks, coding, computer use, and cybersecurity. Notably, M3.1 Flash operates nearly three times faster than its predecessor while reducing operational costs by an impressive 70%. This model builds on the lessons learned from Space Bunny, a previous version, processing over 200 trillion tokens anonymously and showcasing robust performance across multiple benchmarks, including AutomationBench-AA, where it excels in managing complex enterprise workflows. With an impressive score of 76.2 on BankerToolBench and a notable 73.7% on DeepSWE v1.1, M3.1 Flash demonstrates advanced capabilities in multi-step planning, consistency in documentation, and software development tasks. It also ranks first among Chinese models for real-world computer operation challenges, highlighting its strong interaction capabilities within live environments. In cybersecurity, scoring 82% on CyberGym, M3.1 Flash exhibits exemplary skills in code auditing and vulnerability management, making it a valuable tool for security professionals. The model is accessible via the MiniMax API and is set to redefine performance standards in AI-driven agentic tasks at a competitive cost.