ITG NowHome › AI glossary › GPQA Diamond
What is GPQA Diamond?
GPQA Diamond is a set of very hard science questions that tests whether an AI can reason like an expert.
- Category Benchmarks
- Also called GPQA
- Updated 20 Oct 2026
GPQA Diamond, explained simply
GPQA Diamond is a test of graduate level questions in biology, physics and chemistry. Experts wrote them to be hard to look up.
The score is the share of questions the model gets right. Higher is better.
This matters because it checks deep reasoning, not just recall. But it is a multiple choice test, so it doesn't show everything.
When many models score high, the test gets worse at telling the best ones apart.
On ITG Now, the AI models page shows GPQA Diamond scores next to other benchmarks, with a link to each source.
Related terms
BenchmarkHumanity's Last ExamReasoning modelBenchmark contaminationLLM
Where you'll see it on ITG Now
- AI modelGemma 4 31B
- AI modelQwen3.8-27B
- AI newsNew US lab gives away a giant AI free
- AI modelClaude Haiku 5.5
- AI modelClaude Opus 5.5
- AI modelDeepSeek-V4.1-Flash
Questions people ask about GPQA Diamond
Is a higher GPQA score better?
Yes. It means the model answered more of the hard science questions correctly.
What does Diamond mean in GPQA Diamond?
Diamond is the hardest, cleanest part of the question set. It keeps only questions experts agreed on.
Does a high GPQA score mean an AI is smarter than a scientist?
No. It is a narrow test of multiple choice questions, not of real research work.