Cursor Composer 2.5: GPT-5.5 Rival, Ultra-Low Cost AI Coding

Cursor has released Composer 2.5, a new AI coding model. This model offers enhanced efficiency and cost-effectiveness for AI programming. Here's the thing: Composer 2.5 performs similarly to top models like Opus 4.7 and GPT-5.5 in tests. What's interesting is its cost per task is less than one dollar. This is significantly lower than competitors, which can cost up to eleven dollars. The model is built on an open-source checkpoint and involved a twenty-five-fold expansion in training tasks. Cursor allocated 85% of its computing resources to additional training. It achieved a 79.8% score on SWE-Bench Multilingual and 63.2% on CursorBench v3.1. The base version costs 50 cents and $2.50 per million input and output tokens respectively. The bottom line: This development could make advanced AI coding tools much more accessible due to its low cost.