← All notes

Summary

Humanity's Last Exam (HLE) is a multi-modal benchmark of 2,500 expert-written questions across dozens of subjects, including mathematics, humanities, and the natural sciences. The authors designed it as a difficult closed-ended academic benchmark, with questions that have known, verifiable answers but cannot be quickly solved by internet retrieval. They note that LLMs now achieve over 90% accuracy on popular benchmarks like MMLU, and that state-of-the-art LLMs demonstrate low accuracy and calibration on HLE. The public release is at lastexam.ai. This note uses the v12 text last revised 19 Oct 2026. [S1]

Claims

• id: C1 claim: 'HLE consists of 2,500 questions across dozens of subjects.' source: S1 evidence: "HLE consists of 2,500 questions across dozens of subjects"
• id: C2 claim: 'The authors say models now achieve over 90% accuracy on popular benchmarks like MMLU.' source: S1 evidence: "LLMs now achieve over 90% accuracy on popular benchmarks like MMLU"
• id: C3 claim: 'They publicly release HLE at lastexam.ai.' source: S1 evidence: "we publicly release HLE at https://lastexam.ai"

Verification

• Code available: The public release is https://lastexam.ai, which returned HTTP 200 on 2026-11-11. No official GitHub URL was named as the primary release. Not run.
• Reproduced: not reproduced (2026-11-11). Code not run.
• Related work: Live note 2601-04301-test-set-contamination-generative-evals. This is the paper-verification note for Humanity's Last Exam itself.
• Limits: This note follows the v12 abstract and key sections (last revised 19 Oct 2026) and does not restate model-by-model scores, which can change across revisions. High accuracy on HLE is described by the authors as evidence on closed-ended academic questions, not as a complete measure of research ability.
• License: CC BY 4.0 on the abs page (2026-11-11; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 24 Jan 2025 (v1); last revised 19 Oct 2026 (v12). Sources: [S1] https://arxiv.org/abs/2501.14249, v12 (19 Oct 2026). checked_on: 2026-11-11.

When to apply

Use when you need a current, expert-written hard academic benchmark and its official public page, not a stale score table.

Facts

• id: F1 claim: 'HLE is a multi-modal benchmark of 2,500 questions across dozens of subjects.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F2 claim: 'The authors report that models already exceed 90% accuracy on popular benchmarks such as MMLU.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F3 claim: 'They publicly release HLE at https://lastexam.ai.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0

Sources

• [S1] arXiv 2501.14249, https://arxiv.org/abs/2501.14249. License: CC BY 4.0. v12 (19 Oct 2026). Accessed 2026-11-11.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.

@misc{iil_note_te7zej23,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{Humanity's Last Exam}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2501-14249-humanitys-last-exam/},
  note = {Permanent ID iil:note:te7zej23. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2501-14249-humanitys-last-exam.md.