Benchmark Almanac

HLE Extended

GPT-5.6 Sol 77.3% leads since August 2026
Not in model cards

Models answer closed-ended frontier academic questions. Mercor adds 447 new expert-written tasks to the 2158 public ones and reports both scores.

Held-out extension of humanitys-last-exam. Mercor shows the original and extended score for each model. The month is approximate. The Wayback Machine first saved the earliest extended pages on August 15th, 2026.

Made by
Mercor
Released
August 2026
Kind
Expert-written questions
Level
Expert
Scored
Checked automatically
Answers
Short answer
Tasks
Public, plus a private test set
Strongest model tested
Claude Opus 5 #10 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
32
Entries on its own leaderboard, 31 models we track

GPT-5.6 Sol leads with 77.3%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

32 entries. Scroll sideways for the rest.

  1. 77.3 GPT-5.6 Sol
  2. 70.7 GPT-5.6 Terra
  3. 61.0 Claude Opus 5
  4. 58.7 Claude Fable 5
  5. 57.5 GPT-5.6 Luna
  6. 56.1 GPT-5.4
  7. 53.3 GPT-6 Luna
  8. 49.4 Claude Opus 4.8
  9. 47.9 Muse Spark Open
  10. 47.1 Qwen 3.8 Max
  11. 45.8 Grok 4.6
  12. 45.3 DeepSeek V4.1 Flash
  13. 36.8 Gemini 3.8 Flash
  14. 36.1 DeepSeek-V4-Flash
  15. 35.0 Kimi K3
  16. 31.6 Gemini 3.7 Flash
  17. 31.2 Claude Opus 4.7
  18. 29.1 Claude Sonnet 5
  19. 25.8 GLM-5.2
  20. 25.4 Grok 4.5
  21. 19.6 Gemini 3.1 Pro
  22. 14.8 Gemini 3.5 Flash
  23. 13.7 Kimi K2.7 Code
  24. 13.3 Gemini 3.6 Flash
  25. 10.9 Claude Sonnet 4.6
  26. 8.10 gpt-oss-120b
  27. 8.10 DeepSeek-V3.2
  28. 6.60 MiniMax-M2.7
  29. 5.40 MiniMax-M3
  30. 4.20 Qwen3.5
  31. 3.80 Gemma 4 31B IT
  32. 2.00 GLM-5
  1. 1 GPT-5.6 Sol max 77.3%
  2. 2 GPT-5.6 Terra max 70.7%
  3. 3 Claude Opus 5 high 61%
  4. 4 Claude Fable 5 high 58.7%
  5. 5 GPT-5.6 Luna max 57.5%
  6. 6 GPT-5.4 xhigh 56.1%
  7. 7 GPT-6 Luna max 53.3%
  8. 8 Claude Opus 4.8 high 49.4%
  9. 9 Muse Spark Open xhigh 47.9%
  10. 10 Qwen 3.8 Max xhigh 47.1%
  11. 11 Grok 4.6 xhigh 45.8%
  12. 12 DeepSeek V4.1 Flash max 45.3%
  13. 13 Gemini 3.8 Flash high 36.8%
  14. 14 DeepSeek-V4-Flash max 36.1%
  15. 15 Kimi K3 max 35%
  16. 16 Gemini 3.7 Flash high 31.6%
  17. 17 Claude Opus 4.7 high 31.2%
  18. 18 Claude Sonnet 5 high 29.1%
  19. 19 GLM-5.2 max 25.8%
  20. 20 Grok 4.5 high 25.4%
  21. 21 Gemini 3.1 Pro high 19.6%
  22. 22 Gemini 3.5 Flash high 14.8%
  23. 23 Kimi K2.7 Code high 13.7%
  24. 24 Gemini 3.6 Flash high 13.3%
  25. 25 Claude Sonnet 4.6 high 10.9%
  26. 26 gpt-oss-120b high 8.10%
  27. 27 DeepSeek-V3.2 8.10%
  28. 28 MiniMax-M2.7 high 6.60%
  29. 29 MiniMax-M3 high 5.40%
  30. 30 Qwen3.5 4.20%
  31. 31 Gemma 4 31B IT 3.80%
  32. 32 GLM-5 2.00%

Source: Official HLE Extended leaderboard

#EntryScoreSettingDateSource
1 GPT-5.6 Sol 77.3% max Jul 2026 Official leaderboard
2 GPT-5.6 Terra 70.7% max Jul 2026 Official leaderboard
3 Claude Opus 5 61% high Jul 2026 Official leaderboard
4 Claude Fable 5 58.7% high Jun 2026 Official leaderboard
5 GPT-5.6 Luna 57.5% max Jul 2026 Official leaderboard
6 GPT-5.4 56.1% xhigh Mar 2026 Official leaderboard
7 GPT-6 Luna 53.3% max Sep 2026 Official leaderboard
8 Claude Opus 4.8 49.4% high May 2026 Official leaderboard
9 Muse Spark Open 47.9% xhigh Oct 2026 Official leaderboard
10 Qwen 3.8 Max 47.1% xhigh Aug 2026 Official leaderboard
11 Grok 4.6 45.8% xhigh Aug 2026 Official leaderboard
12 DeepSeek V4.1 Flash 45.3% max Sep 2026 Official leaderboard
13 Gemini 3.8 Flash 36.8% high Sep 2026 Official leaderboard
14 DeepSeek-V4-Flash 36.1% max Apr 2026 Official leaderboard
15 Kimi K3 35% max Jul 2026 Official leaderboard
16 Gemini 3.7 Flash 31.6% high Aug 2026 Official leaderboard
17 Claude Opus 4.7 31.2% high Apr 2026 Official leaderboard
18 Claude Sonnet 5 29.1% high Jun 2026 Official leaderboard
19 GLM-5.2 25.8% max Jun 2026 Official leaderboard
20 Grok 4.5 25.4% high Jul 2026 Official leaderboard
21 Gemini 3.1 Pro 19.6% high Feb 2026 Official leaderboard
22 Gemini 3.5 Flash 14.8% high May 2026 Official leaderboard
23 Kimi K2.7 Code 13.7% high Jun 2026 Official leaderboard
24 Gemini 3.6 Flash 13.3% high Jul 2026 Official leaderboard
25 Claude Sonnet 4.6 10.9% high Feb 2026 Official leaderboard
26 gpt-oss-120b 8.10% high Aug 2025 Official leaderboard
27 DeepSeek-V3.2 8.10% Dec 2025 Official leaderboard
28 MiniMax-M2.7 6.60% high Mar 2026 Official leaderboard
29 MiniMax-M3 5.40% high Jun 2026 Official leaderboard
30 Qwen3.5 4.20% Official leaderboard
31 Gemma 4 31B IT 3.80% Apr 2026 Official leaderboard
32 GLM-5 2.00% Feb 2026 Official leaderboard

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard 1 leader
GPT-5.6 Sol 77.3%

Source: Official HLE Extended leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Sol Claude Fable 5 GPT-5.4 Gemini 3.1 Pro Claude Sonnet 4.6 gpt-oss-120b 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Sol Claude Fable 5

Source: Official HLE Extended leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Knowledge

All 121 in Knowledge
  • MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects. Intern-S2-397B 89.8%
  • SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy. Qwen 3.6 Max (Preview) 73.9%
  • WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1. Tencent Hy4 preview 83.9%
  • BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly. Gemma 4 31B IT 74.4%
  • KMMLU-Pro In use Models answer multiple-choice questions taken from Korean National Professional Licensure exams. It tests professional knowledge used in Korean industry. Solar Pro 4 79.2%
  • HLE-Verified In use Verified and revised version of Humanity's Last Exam (1,811 items) with uncertain items removed. Seed2.0 Pro 73.6%