Skip to main content
idapt
AccueilModèles IATarifs
Sign inStart free
All terms

Learn

What is Humanity's Last Exam (HLE)?

Last reviewed July 16, 2026

Humanity's Last Exam (HLE) is a benchmark of 2,500 questions at the frontier of human expertise, contributed by subject-matter experts across mathematics, the sciences, and the humanities specifically because current models fail them. It was created as earlier exams saturated (top models clustering near the ceiling) and is designed to stay hard, so scores are low by construction and gaps between models stay visible.

Why saturated benchmarks stop working

Once leading models all score in the high 90s on a test, it stops discriminating: differences shrink to noise and contamination. HLE resets the ceiling with questions hard enough that frontier models miss most of them, restoring a usable signal at the top end. Its numbers look low next to older benchmarks; that is the design working.

How to read HLE scores

Read them comparatively, not absolutely: a model in the twenties is meaningfully ahead of one in the teens even though both fail most questions. Ranking shifts at the top of HLE track frontier progress more sensitively than saturated suites, which is why it features in capability indices despite its youth.

Top models on Humanity's Last Exam

Full leaderboard
  1. 1Muse Spark 1.152.2%
  2. 2Gemini 3.1 Pro46.4%
  3. 3Kimi K343.5%
  4. 4Qwen3.7 Max38.1%
  5. 5Gemini 3 Pro37.5%
  6. 6GPT 5.436.2%
  7. 7Claude Opus 4.634.4%
  8. 8GPT 525.3%
  9. 9Kimi K2.524.4%
  10. 10o320.3%

Peak scores from the compiled benchmark data; reasoning models use their highest effort tier.

In idapt

idapt ranks catalog models on HLE, shows each score in the model page benchmark table, and lets you run the same hard question across several models to see the difference yourself.

HLE leaderboardCompare models

Frequently asked

Related terms

GPQAFrontierMathEpoch Capabilities Index

Multi-Model Chat

GPT, Claude, Gemini, Grok, Llama: pick mid-conversation, cheapest-first.

Web Search

Real-time info, cited sources, every model: no per-provider tax.

See it in practice

The fastest way to understand it is to use it. Start free, no credit card.

Try idapt free
  • Accueil
  • Tarifs
  • Modèles IA
  • Modèles d'image
  • Modèles de voix
  • Modèles vidéo
  • Classements
  • New models
  • Model status
  • Multi-Model Chat
  • Voice Mode
  • Voice HUD
  • Web Search
  • Image Generation
  • Video Generation
  • Audio Generation
  • Transcription
  • Drive
  • Secrets
  • Sharing
  • Workspaces
  • Tasks
  • Memory
  • Agents
  • Subagents
  • Automations
  • Skills
  • Code Execution
  • Computers
  • Computer Use
  • Cloud Computers
  • Local AI
  • AI Gateway
  • API & SDK
  • CLI
  • MCP
  • Tunnels
  • All features →
  • Blog
  • Cas d'usage
  • Comparisons
  • Best of
  • Learn
  • Free tools
  • Prompts
  • Templates
  • Changelog
  • Centre d'aide
  • FAQ
  • Confidentialité
  • Comparer tous les modèles
  • Support
  • Développeurs
  • Quickstarts
  • Référence API
  • Tarifs API
  • CLI
  • MCP
  • Downloads
  • Desktop
  • Badges and embeds
© idapt[email protected]ConditionsPolitique de confidentialitéMentions légalesSignaler un contenu
X (Twitter)