What it measures
- Expert knowledge
- Multimodal question answering
- Frontier-model saturation resistance
A broad expert-level benchmark intended to test difficult questions across many domains and formats.
% accuracy; higher is better.
| Rank | Model | Provider | Score | Source | Source Type | Verified |
|---|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro | 47.31 | Scale Labs HLE Text Only leaderboard Accessed 2026-06-12 | leaderboard | Yes | |
| 2 | Gemini 3 Pro | 37.72 | Scale Labs HLE Text Only leaderboard Accessed 2026-06-12 | leaderboard | Yes | |
| 3 | OpenAI GPT-5.4 | OpenAI | 36.47 | Scale Labs HLE Text Only leaderboard Accessed 2026-06-12 | leaderboard | Yes |
| 4 | Claude Opus 4.6 | Anthropic | 36.24 | Scale Labs HLE Text Only leaderboard Accessed 2026-06-12 | leaderboard | Yes |
| 5 | DeepSeek-R1 | DeepSeek | 8.54 | Scale Labs HLE Text Only leaderboard Accessed 2026-06-12 | leaderboard | Yes |
leaderboard
The paper is static; use the Scale Labs text-only leaderboard for current text-only scores and the official hub for benchmark definition. Text-only leaderboard covers the 86% text-only HLE subset; do not mix it with multimodal or HLE-Rolling scores without labeling.
paper
Use for benchmark definition, not necessarily for latest model scores.
official
Use for dataset, code, or implementation details; score freshness depends on the benchmark.
official
Official benchmark hub for paper, dataset, and methodology links. Use the Scale Labs leaderboard for the text-only operational score table.