Humanity's Last Exam: 2,500 Multi-Modal Frontier Knowledge Questions

Humanity's Last Exam (HLE) is a multi-modal benchmark containing 2,500 questions across dozens of academic subjects, released by the Center for AI Safety and Scale AI in January 2026. It serves as a frontier-level evaluation suite designed to test the limits of human knowledge through closed-ended questions.

Use Cases

• Evaluating Large Language Model (LLM) performance on frontier-level academic questions
• Testing multi-modal reasoning capabilities using the image-based question components
• Benchmarking domain-specific expertise across dozens of specialized subjects

Strengths

• 2,500 expert-level questions
• Multi-modal support including both text and image modalities
• Broad subject coverage across dozens of academic fields
• MIT licensed for research use

Limitations

• Small sample size of 2,500 records limits its use for model training
• Closed-ended format may not evaluate open-ended generative reasoning
• High risk of benchmark contamination if redistributed against author requests

Provenance

Source

Center for AI Safety & Scale AI

Freshness

Last updated January 20, 2026.

Geography

United States