Learn
What is SimpleQA?
Last reviewed July 16, 2026
SimpleQA is a factuality benchmark made of short, unambiguous questions with exactly one verifiable answer, graded as correct, incorrect, or not attempted. Because questions are simple to check and hard to bluff, it measures hallucination about as directly as any public test: a model that answers confidently but wrongly scores worse than one that declines. Frontier models miss a large share, which calibrates how much to trust unaided recall.
What the grading rewards
Three outcomes per question (correct, incorrect, not attempted) let the benchmark separate knowledge from calibration: answering only when sure beats guessing. A model can raise its score by declining questions it would get wrong, which is exactly the behavior you want from a model reporting facts.
Why scores run low
The questions target long-tail facts (dates, names, niche specifics) where training data is thin, so unaided models miss a large share. The practical lesson is not that models are useless for facts: it is that grounding (web search, retrieval, citations) does the factual heavy lifting, and SimpleQA quantifies what recall alone is worth.
Top models on SimpleQA
Full leaderboardPeak scores from the compiled benchmark data; reasoning models use their highest effort tier.
idapt ranks models on SimpleQA in the factuality leaderboard, and for real fact-finding any chat can ground itself with web search and run the same question across models to catch disagreement.
Frequently asked
See it in practice
The fastest way to understand it is to use it. Start free, no credit card.
Try idapt free