Cursor has shipped Composer 2, its updated agentic coding feature, with a stated focus on frontier-level performance on hard coding tasks.
Composer is Cursor’s multi-file editing agent — the mode where you describe what you want built and the AI writes, edits, and coordinates changes across your codebase. Version 2 targets the tasks where earlier versions stumbled: complex, multi-step problems that require the model to reason across files and maintain coherent intent over longer sessions.
The changelog is light on specific benchmark numbers, but the framing — “frontier-level coding performance” and “strong results on challenging coding tasks” — signals Cursor is positioning Composer 2 directly against the top-tier coding models rather than treating it as a convenience feature.
If you use Cursor for anything beyond single-file edits, this is worth testing on your hardest recurring tasks. Open Composer, throw a problem at it that previously required heavy back-and-forth, and see where it breaks down now versus before.