// benchmark explainer
Humanity's Last Exam (HLE), explained
By the TrueIQ editorial team at Ahzy Capital · Last reviewed Oct 15, 2026 · How we work
Humanity's Last Exam is a large set of extremely hard questions written by subject experts around the world, designed to stay difficult for AI after older benchmarks saturated.
| Made by | Center for AI Safety and Scale AI, with expert contributors (2025) |
|---|---|
| Format | Multiple choice and exact answer questions; some include images |
| Subjects | Maths, sciences, humanities and more |
| Paper | arXiv 2501.14249 |
What it measures
Each question was checked to make sure leading models at the time couldn't answer it, then reviewed by experts. A large share is maths. The aim is to measure frontier academic knowledge and reasoning, not everyday usefulness.
Where it falls short
- Because questions were filtered against the models of the time, early scores were low partly by design.
- Some questions have been disputed, and the authors have issued corrections.
- Results differ a lot depending on whether tools like web search or code execution were allowed.
How to read a score
Always check whether a score was run with or without tools. A 'with search' score isn't comparable with a 'no tools' score.
Where to see current scores
We don't republish scores. Leaderboards change weekly, and numbers depend on how the test was run. Check these instead:
- Artificial Analysis: Humanity's Last Exam (HLE) results (independently run)
- HLE official site
- HLE paper