Mercor HLE Extended - Pass@1: leaderboard

Mercor's held-out extension of Humanity's Last Exam (HLE), the closed-ended frontier academic question set: models answer 447 Mercor questions kept apart from HLE's 2,158 public ones, and the score is Pass@1 (%), one row per model setting on Mercor's leaderboard. Mercor prints most models' original HLE score beside it; that score is not used here, and the page does not say how answers are graded.

Metric: Pass@1 (%). Source: www.mercor.com. Saturation forecast: Around February 2027. 32 models tracked. Updated 19 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.6 Pro SolMax77.3#31
2GPT-5.6 TerraMax70.7#72
3Claude Opus 5High61#26
4Claude Fable 5High58.7#32
5GPT-5.6 LunaMax57.5#133
6GPT-5.4xHigh56.1#138
7GPT-6 LunaMax53.3#106
8Claude Opus 4.8High49.4#64
9Muse Spark OpenxHigh47.9#82
10Qwen 3.8 MaxxHigh47.1#66
11Grok 4.6xHigh45.8#50
12DeepSeek V4.1 FlashMax45.3#69
13Gemini 3.8 FlashHigh36.8#43
14DeepSeek V4 FlashMax36.1#267
15Kimi K3Max35#61

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mercor-hle-extended-pass-1 · How It Works · Data refreshed daily, snapshot 2026-11-20.