MiniMax M2.5 has launched as the first open-weight AI model to achieve over 80% on the SWE-Bench Verified, scoring 80.2%, which evaluates bug fixes in real GitHub repositories. This model is specifically tuned for agentic tasks such as coding, searching, and creating office files, making it particularly effective for enterprise use. Notably, it drops the cost of frontier AI performance by up to 95%, with an input price of $0.30/M and output at $1.20/M, while providing similar performance to Opus 4.6 and GPT 5.2. This significant cost reduction facilitates scalable deployments of long-horizon agents for various tasks.
MiniMax M2.5 becomes first open-weight model to score 80% on SWE-Bench Verified
