Cursor Self-Developed Composer 2 Release: Programming Benchmark Beyond Opus 4.6, Price Reduced by 14% Compared to Previous Generation

According to 1M AI News monitoring, the AI programming tool Cursor has released the third-generation proprietary programming model Composer 2. The standard version is priced at $0.50/$2.50 (input/output per million tokens), representing a price reduction of around 86% compared to the previous generation Composer 1.5. Additionally, a faster version with equivalent intelligence (priced at $1.50/$7.50) has been introduced as the default option.

Results from benchmarks published by Cursor show that Composer 2 scored 61.7% on the programming Agent benchmark Terminal-Bench 2.0, surpassing Opus 4.6 at 58.0% but lagging behind GPT-5.4 at 75.1%. In the software engineering benchmark SWE-bench Multilingual, it scored 73.7% (compared to the previous generation's 65.9%), while Cursor's proprietary CursorBench scored 61.3% (compared to the previous generation's 44.2%). Cursor stated that the performance improvement comes from the initial continued pretraining of the base model, followed by reinforcement learning training to complete long-cycle programming tasks requiring hundreds of steps.

This model is only available within Cursor and is not offered as a standalone API. Cursor's parent company Anysphere is currently valued at $293 billion.