MiniMax M2.1 has launched on Kilo, following the success of its predecessor, M2, which ranked third on Kilo’s AI coding models leaderboard. This new model is already outperforming competitors like DeepSeek and Kimi on key benchmarks. For instance, M2.1 achieved a score of 74.0 on SWE-bench Verified, surpassing DeepSeek V3.2 and Kimi K2. It also scored 91.5 on the VIBE-Web benchmark, which evaluates web app functionality, outpacing the Claude Opus and Claude Sonnet models. In real-world testing, M2.1 demonstrated impressive performance across various scenarios. The Kilo team tasked it with building an Express API, debugging a React game, extending a Go file-serving app, and documenting a Rust CLI. M2.1 excelled in all these areas, showing significant improvements in speed and efficiency compared to M2. Users can access M2.1 through Kilo Code without any downloads or GPU setup, making it accessible for quick development. For those unsure how to leverage M2.1, Kilo Code offers a Code Review feature. This AI-powered tool analyzes pull requests in real-time, providing structured feedback directly on GitHub. Using M2.1 as the reviewing model can enhance its performance in production settings.