Cursor has released Composer 2, its third-generation coding model that outperforms Anthropic’s Opus 4.6 on key benchmarks while being significantly cheaper. The new model costs as little as $0.5 per million input tokens and $2.5 per million output tokens, making it a more affordable option for developers.
Compared to Opus 4.6, Composer 2 scores higher on Terminal-Bench 2.0, a benchmark that measures AI agents’ performance in real-world software engineering tasks. While still behind OpenAI’s GPT-5.4, the gap between Cursor’s models has narrowed significantly with each new release.
Cursor’s approach to model development involves continuous pre-training and a unique “self-summarization” technique that compresses the model’s memory. This innovation reduces compaction errors by 50% compared to traditional methods.
The company’s Auto mode allows developers to choose the best model based on a trade-off between intelligence, speed, and cost, making it easier for them to select the most suitable option for their projects. With Composer 2, Cursor has taken another step towards becoming a leader in AI coding solutions.
Source: https://thenewstack.io/cursors-composer-2-beats-opus-46-on-coding-benchmarks-at-a-fraction-of-the-price