GPQA Diamond
A difficult subset of GPQA for graduate-level science question answering evaluation.
What this benchmark measures
GPQA Diamond focuses on difficult graduate-level science questions. Evaluations typically report accuracy on a fixed subset with a specified prompt format and answer extraction method.
Rows on this page come from public model-system reports rather than a single central GPQA leaderboard. Treat the prompt format, answer extraction, and reasoning policy as part of the score context.
The metric shown here is benchmark-local accuracy. It should be compared within GPQA Diamond only.
What to be careful about
Prompt format, answer extraction, and subset selection can materially change reported accuracy.
Frequently asked
What is GPQA Diamond?
A difficult subset of GPQA for graduate-level science question answering evaluation. It is a reasoning benchmark measured by accuracy.
What does accuracy mean on GPQA Diamond?
GPQA Diamond reports accuracy (%); higher is better. Scores are shown only within GPQA Diamond and are never averaged with other benchmarks.
What is the top reported GPQA Diamond score?
Fugu has the top reported score on GPQA Diamond: 95.5% (accuracy).
Why do GPQA Diamond scores differ across runs?
Harness, scaffold, reasoning effort, and prompt setup change results, so two runs of the same model can differ. evals.report keeps each score with its run context so the differences stay visible.
Does evals.report rank models across benchmarks?
No. GPQA Diamond scores are shown within their own metric; evals.report never combines benchmarks into a composite ranking or a single "best model".