Cursor has launched Composer, its first proprietary coding model, alongside Cursor 2.0, a major platform update focused on multi-agent development and faster, more autonomous workflows.
Key updates
• Composer model: A mixture-of-experts (MoE) system trained on large, real-world codebases. It delivers near-frontier reasoning and generates code up to 4× faster than comparable tools, completing most requests in under 30 seconds.
• Agent-first design: Cursor 2.0 shifts from a file-centric IDE to an agent-centric workspace, letting developers run up to eight AI assistants in parallel, independently or on the same problem for best-of-many results.
• Built-in browser: Agents can now test and iterate on their output directly inside Cursor, with isolated environments for safety.
• Voice and real-time review: Users can give feedback, highlight code for fixes, and switch seamlessly between the classic IDE and the new Agent View.
• Fully in-house: Composer is Cursor’s first homegrown model, marking a step away from dependence on third-party LLMs.
Why it matters
It’s a glimpse into a workflow where human developers focus less on typing and more on judgment, architecture, and quality.
Try it now: https://cursor.com/download