- Cleus
- AI benchmarks
- CursorBench
CursorBenchexplained
Real multi-file coding jobs inside the Cursor editor. CursorBench is Cursor’s own test for AI coding agents, built from the kind of vague, multi-file tasks people hand to the agent in the Cursor code editor. The agent works inside Cursor’s own setup, and its results are graded on whether the change is correct, the quality of the code and how efficiently it got there.
- Highest published score
- Claude Opus 5.5, 57.8%
- Leader among popular models
- Claude Opus 5.5
- Models with a result
- 19
- Skill it tests
- Coding
What CursorBench really tests
- What it is
- CursorBench is Cursor’s own test for AI coding agents, built from the kind of vague, multi-file tasks people hand to the agent in the Cursor code editor. The agent works inside Cursor’s own setup, and its results are graded on whether the change is correct, the quality of the code and how efficiently it got there.
- A question like
- “The settings page sometimes saves the old value — find out why and fix it.” The agent has to explore the codebase, find the bug across several files and make a fix that passes review. Our own example in the style of the test, not a real item.
- Why it matters
- It is close to how developers actually use AI day to day: messy instructions, real codebases, no hand-holding.
- Reading the score
- Share of tasks completed to Cursor’s standard.
- The catch
- Cursor builds, runs and scores it, so it isn’t independent, and scores include Cursor’s own agent tooling, not just the model. Versions change; only compare scores from the same version.
- Made by
- Cursor (Anysphere), 2026. Official page
Leaderboard
Each model’s best published result across its reasoning settings.
Popular models
The models we track across this site.
Top 19 of all 19 models tested
Includes older, preview and specialist models that aren’t on our main leaderboard.
Questions
What is CursorBench?
CursorBench is Cursor’s own test for AI coding agents, built from the kind of vague, multi-file tasks people hand to the agent in the Cursor code editor. The agent works inside Cursor’s own setup, and its results are graded on whether the change is correct, the quality of the code and how efficiently it got there.
What does a CursorBench score mean?
Share of tasks completed to Cursor’s standard.
Which AI model is best on CursorBench?
Claude Opus 5.5 from Anthropic has the highest published CursorBench result we have (57.8%).
Is CursorBench a good benchmark?
It is close to how developers actually use AI day to day: messy instructions, real codebases, no hand-holding. The catch: Cursor builds, runs and scores it, so it isn’t independent, and scores include Cursor’s own agent tooling, not just the model. Versions change; only compare scores from the same version.
Who made CursorBench?
Cursor (Anysphere), 2026.
Benchmark scores: Epoch AI, Capabilities & benchmarking, used under CC BY 4.0. For each model we show its best published result across reasoning settings. Prices and context sizes: OpenRouter. Data refreshed November 9, 2026. Hands-on tests (answers, websites, code, refusals, speed) are our own single runs. Explanations written by Cleus.