Cursor has introduced a new artificial intelligence model, Composer 2.5. According to the company, the model is comparable in performance to OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7, but costs about 10 times less to run.
The Composer 2.5 model is built on Kimi 2.5, developed by Chinese-based Moonshot AI. Cursor says that additional training used 85% more computing resources than the creators of the base model.
According to the company, the new model performs better on long-term tasks and in large contextual memory. Composer 2.5 was trained using the "self-distillation" method. In this approach, the same model acts as both a "teacher" and a "student."
According to test results, Composer 2.5 outperformed the previous Composer 2 model in the Terminal-Bench 2.0, SWE-Bench Multilingual and CursorBench v3.1 benchmarks, and showed results at the level of GPT-5.5 and Claude Opus 4.7.
In terms of price, the model is noteworthy:
1 million input tokens — $ 0.5
1 million output tokens — $ 2.5
For comparison:
GPT-5.5 — $ 5 and $ 30, respectively
Claude Opus 4.7 — $ 5 and $ 25
Cursor also said that it is working on a new model from scratch together with xAI, and this project uses “10 times more” computing power.
