SimpleQA

SimpleQA: Measuring short-form factuality in large language models

4,326 short fact-seeking questions with a single indisputable answer, adversarially collected against GPT-4o and verified by two independent annotators, spanning science, history, art, geography and more. A grader model marks each answer correct, incorrect or not attempted; the headline metric is percent correct, with F-score and calibration also reported. It measures whether models know what they know: hallucination-prone models score low even when otherwise capable.

62.5%
gpt-4.5-preview-2025-02-27
self-reported
2024-11-07 → 2026-08-13: 38.2% → 57.9%
Released
2024-11
Maintainer
OpenAI (Wei et al.)
Status
active
Contamination
high
Metric
accuracy (correct) (percent, ↑)
Tasks
4,326
Domains
factuality knowledge
human
no measured baseline

Notes. Answers are static facts, so retrieval or web search makes the test trivial; only no-tools runs are comparable. Google's SimpleQA Verified (1,000 relabeled questions, 2025) is a separate, stricter variant and its scores should not be mixed with the original. OpenAI stopped reporting SimpleQA in simple-evals in July 2025.

Full ledger

SystemDeveloperScoreDateSourceConditions
gpt-4.5-preview-2025-02-27OpenAI62.5%self-reportedtools: no
simple-evals README benchmark table (repo deprecated July 2025).
DeepSeek V4 Pro 0813DeepSeek57.9%aggregator
benchlm.ai lists this under SimpleQA, but DeepSeek's own model card reports 57.9 as SimpleQA-Verified (Pass@1), a different 1,000-question variant; treat as not comparable with the rows above.
o3OpenAI49.4%self-reportedtools: no
simple-evals README; o3-high 48.6, o4-mini 20.2.
GPT-4oOpenAI38.2%papertools: no
Table 3 of the SimpleQA paper: percent correct; o1-preview scored 42.7.