Reasoning leaderboard
GPQA Diamond
Graduate-level science questions that resist search.
Scores are temporarily unavailable. Please check back shortly.
What it measures
The hardest subset of GPQA: multiple-choice questions in biology, chemistry and physics written by PhD-level domain experts, selected so that skilled non-experts with web access still get them wrong.
Beyond public benchmarks
Get a private eval built for your model.
Frontier models are closing in on GPQA’s ceiling. Separating them now takes fresh, unpublished questions written and verified by working scientists in your target domains.
- Unpublished tasks written by vetted US-based subject-matter experts
- Rubric-based grading with expert review, not just automated scoring
- Delivered under NDA, with training data to close the gaps it finds
More leaderboards
Agentic
Tau-Bench Airline
Search
BrowseComp
Search
Humanity’s Last Exam
Search
DeepSearchQA
Search
WideSearch
Design Arena / App Building
Full-Stack Apps
Design Arena / App Building
Web Apps
Design Arena / App Building
Mobile Apps
Design Arena / App Building
Native Android
Design Arena / App Building
Godot Game Dev
Design Arena / Design
Websites
Design Arena / Design
UI Components
Design Arena / Design
Game Dev
Design Arena / Design
3D
Design Arena / Design
Data Visualization
Design Arena / Design