CursorBench 4.0

Cursor's own test of coding agents on ambiguous, multi-file tasks taken from real Cursor sessions — editing, refactoring, investigating a codebase, understanding what the user meant, managing jobs and following a design. Cursor runs each model at several reasoning efforts; each release here carries the score of its best listed effort. Scores aren't comparable with earlier CursorBench versions. Higher is better.

Claude Opus 5.5 by Anthropic holds the best CursorBench 4.0 result among tracked models, at 57.8% (released Sep 22 2026). Higher scores are better on this benchmark.

1. Claude Opus 5.5 (Anthropic) — 57.8%; 2. Claude Sonnet 5.5 (Anthropic) — 55.5%; 3. Claude Fable 5.1 (Anthropic) — 51.8%; 4. Claude Opus 5 (Anthropic) — 46.6%; 5. Grok 4.7 (SpaceXAI) — 46.3%.

13 tracked models have a published CursorBench 4.0 score. Scores are gathered from CursorBench (Cursor), which runs the benchmark and publishes the full field, and are kept in step with its board.