Cursor ships Composer 2.5 model matching Opus and GPT-5.5

cursor released composer 2.5, an upgrade built on the open-source kim k2.5 checkpoint, allocating 85 percent of compute to reinforcement learning on 25 times more synthetic tasks. the model rivals top-tier proprietary systems, scoring 79.8 percent on swe-bench multilingual and 63.2 percent on cursorbench v3.1. cursor highlights extreme cost efficiency of $0.50 per million input tokens and $2.50 per million output tokens, undercutting major lab prices. the company is already developing a larger successor using the colossus-2 cluster with spacex and xai.