- Cleus
- AI benchmarks
- GPQA Diamond
GPQA Diamondexplained
PhD-level science questions you can’t just Google. GPQA Diamond is 198 multiple-choice questions in biology, physics and chemistry, written and checked by people with PhDs in those fields. The questions were picked so that experts get them right but smart non-experts don’t, even with unlimited time on the web.
- Highest published score
- Gemini 4 Argon, 96.4%
- Leader among popular models
- Gemini 4 Argon
- Models with a result
- 227
- Skill it tests
- Reasoning
What GPQA Diamond really tests
- What it is
- GPQA Diamond is 198 multiple-choice questions in biology, physics and chemistry, written and checked by people with PhDs in those fields. The questions were picked so that experts get them right but smart non-experts don’t, even with unlimited time on the web.
- A question like
- A chemistry question describes a reaction sequence in three steps and asks which of four products forms at the end. Getting it right means actually working through the chemistry, not recalling a fact. Our own example in the style of the test, not a real item.
- Why it matters
- It is the most widely quoted test of whether a model can reason through expert-level science instead of reciting what it read. If you use AI for studying, research or technical work, this is the closest thing to “does it understand the subject”.
- Reading the score
- Share of the 198 questions answered correctly. Random guessing scores 25%. In the original study, PhD experts scored about 65% in their own field; skilled non-experts with web access scored about 34%.
- The catch
- It is multiple choice with four options, so guessing gives 25%. The best models are now above 90%, so it no longer separates the top models well. Small differences near the top are mostly noise.
- Made by
- Rein et al. (NYU, Cohere, Anthropic), 2023. Official page
Leaderboard
Each model’s best published result across its reasoning settings.
Popular models
The models we track across this site.
Top 20 of all 227 models tested
Includes older, preview and specialist models that aren’t on our main leaderboard.
Questions
What is GPQA Diamond?
GPQA Diamond is 198 multiple-choice questions in biology, physics and chemistry, written and checked by people with PhDs in those fields. The questions were picked so that experts get them right but smart non-experts don’t, even with unlimited time on the web.
What does a GPQA Diamond score mean?
Share of the 198 questions answered correctly. Random guessing scores 25%. In the original study, PhD experts scored about 65% in their own field; skilled non-experts with web access scored about 34%.
Which AI model is best on GPQA Diamond?
Gemini 4 Argon from Google DeepMind has the highest published GPQA Diamond result we have (96.4%).
Is GPQA Diamond a good benchmark?
It is the most widely quoted test of whether a model can reason through expert-level science instead of reciting what it read. If you use AI for studying, research or technical work, this is the closest thing to “does it understand the subject”. The catch: It is multiple choice with four options, so guessing gives 25%. The best models are now above 90%, so it no longer separates the top models well. Small differences near the top are mostly noise.
Who made GPQA Diamond?
Rein et al. (NYU, Cohere, Anthropic), 2023.
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.