Cursor introduces its coding model alongside multi-agent interface

Cursor has made a significant leap in the coding landscape by launching what it claims to be a competitive coding model, along with an upgraded version 2.0 of its integrated development environment (IDE). This new iteration introduces a remarkable feature that enables the execution of tasks through multiple agents simultaneously. The company’s flagship IDE draws inspiration from Visual Studio Code while placing a strong focus on vibe coding and seamless integration of large language model-based tools into both its interface and workflow. Since its inception, Cursor has showcased support for models created by other industry giants, including OpenAI, Google, and Anthropic. Although the company has previously tested its own built-in models, they did not measure up to the leading frontier models in the market. Now, Cursor asserts that its newly developed model, Composer, changes the game. Engineered with reinforcement learning and a mixture-of-experts architecture, Composer is confidently branded as a ‘frontier model’ that boasts a speed increase of four times compared to other similarly intelligent models. This claim is particularly striking, considering the competitive landscape. When examining performance metrics, Cursor’s internal benchmark, Cursor-Bench, reveals that while Composer may lag behind the 'best frontier' models in terms of overall intelligence and tokens processed per second, it excels in speed, significantly outpacing not only its direct competitors but also leading open models and speed-focused frontier technologies.