- Cleus
- AI benchmarks
- Humanity’s Last Exam
Humanity’s Last Examexplained
Thousands of the hardest questions experts could write. Humanity’s Last Exam (HLE) is about 2,500 questions written by around 1,000 subject experts across more than 100 fields, from classics to quantum physics. Each question was kept only if the leading AI models of the time got it wrong. Some questions include images.
- Highest published score
- GPT-6 Astra, 54.8%
- Leader among popular models
- GPT-6 Astra
- Models with a result
- 46
- Skill it tests
- Reasoning
What Humanity’s Last Exam really tests
- What it is
- Humanity’s Last Exam (HLE) is about 2,500 questions written by around 1,000 subject experts across more than 100 fields, from classics to quantum physics. Each question was kept only if the leading AI models of the time got it wrong. Some questions include images.
- A question like
- A question shows a short passage in an ancient script and asks for a translation of one specific word, or asks for the exact answer to a niche maths or ecology problem with a single correct number. Our own example in the style of the test, not a real item.
- Why it matters
- It was built to be the hardest closed-book academic test around, so it still shows clear gaps between models where older tests are maxed out.
- Reading the score
- Share of questions answered exactly right.
- The catch
- Reviewers have found that a share of the answers are disputed or wrong, which caps how high anyone can score. Results are only available for some models, and they depend on whether tools like web search were allowed.
- Made by
- Center for AI Safety and Scale AI, 2025. 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 46 models tested
Includes older, preview and specialist models that aren’t on our main leaderboard.
Questions
What is Humanity’s Last Exam?
Humanity’s Last Exam (HLE) is about 2,500 questions written by around 1,000 subject experts across more than 100 fields, from classics to quantum physics. Each question was kept only if the leading AI models of the time got it wrong. Some questions include images.
What does a Humanity’s Last Exam score mean?
Share of questions answered exactly right.
Which AI model is best on Humanity’s Last Exam?
GPT-6 Astra from OpenAI has the highest published Humanity’s Last Exam result we have (54.8%).
Is Humanity’s Last Exam a good benchmark?
It was built to be the hardest closed-book academic test around, so it still shows clear gaps between models where older tests are maxed out. The catch: Reviewers have found that a share of the answers are disputed or wrong, which caps how high anyone can score. Results are only available for some models, and they depend on whether tools like web search were allowed.
Who made Humanity’s Last Exam?
Center for AI Safety and Scale AI, 2025.
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.