Cursor 2.0: Composer and new UX in 12 Minutes

Quick Overview

Cursor 2.0 introduces the new Composer agent model, which achieves frontier coding results with generation speeds up to four times faster than similar models like GPT-5 and Sonnet 4.5, while also featuring a dedicated multi-agent interface, built-in browser functionality, and improved UX for interactive coding.

Key Points: Composer is Cursor's new agent model designed for software engineering, achieving frontier coding results on benchmarks. Composer generates code up to four times faster than similar models like GPT-5 and Sonnet 4.5. Cursor 2.0 introduces a Multi-Agent Interface allowing users to run multiple models in parallel, locally or in the cloud. The new version includes a built-in browser tab accessible via the agent interface for navigation and instruction. The model was trained using reinforcement learning (RL) on real-world software engineering challenges, including access to production search/editing tools. Composer demonstrates improved software engineering ability over time with more training compute, consistently increasing its Cursor Bench score. The new UX features include a dedicated view for agent interactions, a unified review interface, and improved developer workflows.

Context: This video provides an overview and demonstration of Cursor 2.0, highlighting its major new feature: Composer, which is described as Cursor's first agent model specifically designed for software engineering tasks, trained using reinforcement learning (RL) to balance intelligence and speed.

Detailed Analysis

The video announces Cursor 2.0, centered around the introduction of the Composer agent model, which is specialized for software engineering, balancing intelligence and speed. Benchmarks show Composer achieving frontier coding results, generating code up to four times faster than models like GPT-5 and Sonnet 4.5. Composer's training involved reinforcement learning (RL) on real-world software engineering challenges, giving it access to production search and editing tools. The presentation includes a benchmark comparison chart illustrating Composer's superior performance in both intelligence and speed metrics. A key new feature is the Multi-Agent Interface, enabling users to run multiple models concurrently, locally or in the cloud. Furthermore, Cursor 2.0 integrates a built-in browser tab directly within the agent panel, allowing users to interactively inspect the rendered output, check console logs, and select DOM elements for context feeding to the agent. The demonstration shows Composer quickly setting up a Next.js blog with a Brutalist theme, handling complex coding tasks, fixing accessibility issues (like contrast ratios), and running commands like directly through the agent interface. The speed of the AI in generating and iterating on code changes, even when dealing with complex requests like styling and accessibility fixes, is emphasized as a major improvement.