BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases8

HLE-Diamond (AA)

As of 28 Oct 2026, the top score on HLE-Diamond (AA) is 67.9% by Claude Opus 5.5 (Anthropic), ahead of Claude Opus 5.5 at 64.0%, across 132 model configurations with a published result (Artificial Analysis, result dated 22 Sept 2026). HLE-Diamond: a 1,000-question refinement of Humanity's Last Exam (500 reasoning, 500 knowledge) used in Artificial Analysis' Intelligence Index v5. HLE-Diamond (AA) counts toward the BenchLeader reasoning score.

Key facts (as of 28 Oct 2026)

  • Claude Opus 5.5 holds the top HLE-Diamond (AA) score at 67.9% (Artificial Analysis, 22 Sept 2026).
  • Claude Opus 5.5 is second on HLE-Diamond (AA) at 64.0% (Artificial Analysis, 22 Sept 2026).
  • HLE-Diamond (AA): Refined 1,000-question HLE (500 reasoning, 500 knowledge) used in Intelligence Index v5. Run by Artificial Analysis (Artificial Analysis).
  • 132 model configurations have a published HLE-Diamond (AA) result on BenchLeader (BenchLeader, 28 Oct 2026).
  • HLE-Diamond (AA) feeds the reasoning category of the BenchLeader Index (BenchLeader methodology).

Refined 1,000-question HLE (500 reasoning, 500 knowledge) used in Intelligence Index v5. Run by Artificial Analysis.

Published by
Artificial Analysis
Category
Reasoning
Index weight
1.0
Models
132
Data as of
28 Oct 2026
Newest result
26 Oct 2026

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

What the test looks like

HLE-Diamond: a 1,000-question refinement of Humanity's Last Exam (500 reasoning, 500 knowledge) used in Artificial Analysis' Intelligence Index v5.

How it is scored

Accuracy percentage on the Diamond subset. Run by Artificial Analysis.

What to keep in mind

Private subset of HLE; not the same as the public HLE board Epoch and Artificial Analysis also publish. Distinct enough from Epoch's full HLE that both count in the BenchLeader Index.

Top 15
  1. 1Claude Opus 5.5 (max)67.9%
  2. 2Claude Opus 5.5 (xhigh)64.0%
  3. 3GPT-6 Astra (max)63.8%
  4. 4GPT-6 Astra (xhigh)61.5%
  5. 5Claude Fable 5.1 (max)60.1%
  6. 6GPT-6 Astra (high)59.3%
  7. 7GPT-6.1 Sol (max)58.8%
  8. 8Claude Fable 5.1 (xhigh)58.8%
  9. 9Gemini 4 Argon (high)56.7%
  10. 10GPT-6.1 Sol (xhigh)55.8%
  11. 11Claude Opus 5.5 (high)55.6%
  12. 12Claude Sonnet 5.5 (max)53.3%
  13. 13GPT-6 Astra (medium)52.7%
  14. 14GPT-6.1 Sol (high)51.9%
  15. 15Claude Opus 5.5 (medium)51.4%
132 of 132
#
1Claude Opus 5.5max67.9%69.82026-09-22
2Claude Opus 5.5xhigh64.0%66.82026-09-22
3GPT-6 Astramax63.8%68.32026-09-03
4GPT-6 Astraxhigh61.5%67.12026-09-03
5Claude Fable 5.1max60.1%66.92026-09-01
6GPT-6 Astrahigh59.3%67.32026-09-03
7GPT-6.1 Solmax58.8%67.62026-09-29
8Claude Fable 5.1xhigh58.8%65.02026-09-01
9Gemini 4 Argonhigh56.7%66.02026-09-30
10GPT-6.1 Solxhigh55.8%64.82026-09-29
11Claude Opus 5.5high55.6%65.92026-09-22
12Claude Sonnet 5.5max53.3%65.72026-09-28
13GPT-6 Astramedium52.7%65.22026-09-03
14GPT-6.1 Solhigh51.9%64.32026-09-29
15Claude Opus 5.5medium51.4%62.22026-09-22
16Claude Fable 5.1high50.4%63.52026-09-01
17Muse Wonder Fullmax48.5%62.72026-10-12
18NewGPT-6.1 Terramax47.3%65.02026-10-26
19GPT-6.1 Solmedium46.0%63.12026-09-29
20GPT-6 Solmax45.2%63.62026-09-22
21GPT-6 Astralow45.1%62.02026-09-03
22Claude Fable 5.1medium45.0%61.52026-09-01
23Claude Sonnet 5.5xhigh44.1%63.42026-09-28
24NewGPT-6.1 Terraxhigh42.3%60.42026-10-26
25Claude Opus 5.5low41.7%57.82026-09-22
26Muse Wondermax39.0%61.22026-10-12
27GPT-6.1 Sollow38.2%59.02026-09-29
28Claude Fable 5.1low37.4%58.12026-09-01
29NewKimi K3.1 Pro37.3%59.82026-10-23
30NewGPT-6.1 Terrahigh35.4%59.92026-10-26
31Claude Sonnet 5.5high33.8%57.92026-09-28
32Muse Spark 1.3max33.7%60.92026-09-02
33Gemini 3.8 Flashhigh33.1%60.32026-09-02
34Gemini 3.1 Pro31.4%60.62026-02-19
35NewGPT-6.1 Terramedium31.3%58.32026-10-26
36Muse Spark 1.3xhigh30.3%58.52026-09-02
37Claude Haiku 5.5max29.6%56.42026-10-07
38Gemini 3.8 Flashmedium29.1%58.62026-09-02
39GPT-5.6 Terramax28.8%61.22026-07-09
40Claude Haiku 5.5xhigh28.8%56.02026-10-07
41MiniMax M3.1 Flash28.1%52.62026-10-19
42Kimi K3max26.9%62.22026-07-16
43Muse Spark Openxhigh26.3%51.72026-10-15
44DeepSeek V4.1 Flash 1017max25.3%57.82026-10-17
45NewGPT-6.1 Terralow24.9%54.92026-10-26
46GPT-5.6 Terraxhigh24.7%60.42026-07-09
47Step 5 Preview24.3%55.02026-09-18
48Gemini 3.8 Flashlow24.3%53.72026-09-02
49Qwen3.8 Maxmax24.0%59.12026-09-02
50Claude Sonnet 5.5medium23.8%50.12026-09-28
51Claude Haiku 5.5high23.7%54.52026-10-07
52Grok 4.7high23.6%56.12026-09-21
53Grok 4.7xhigh23.2%57.42026-09-21
54GPT-6 Lunamax22.2%55.42026-09-22
55Deepseek v4 Flash Visionmax21.1%–2026-08-21
56GPT-5.6 Terrahigh21.0%57.02026-07-09
57Claude Sonnet 5.5low20.6%49.12026-09-28
58DeepSeek V4 Promax19.5%54.52026-08-13
59DeepSeek V4.1 Flashmax19.1%58.92026-09-10
60GPT-6 Lunaxhigh18.9%50.22026-09-22

More reasoning benchmarks · All LLM benchmarks

Cite as: BenchLeader, “HLE-Diamond (AA) leaderboard”, data as of 28 Oct 2026, https://www.benchleader.com/benchmarks/aa_hle_diamond.

HLE-Diamond (AA): questions

What does HLE-Diamond (AA) measure?
HLE-Diamond: a 1,000-question refinement of Humanity's Last Exam (500 reasoning, 500 knowledge) used in Artificial Analysis' Intelligence Index v5. Scores are reported in percent of tasks solved; higher is better.
What is the top score on HLE-Diamond (AA)?
As of 28 Oct 2026, the highest score on HLE-Diamond (AA) is 67.9% by Claude Opus 5.5 (result dated 22 Sept 2026), ahead of Claude Opus 5.5 at 64.0%. Scores come from Artificial Analysis and are re-read daily.
Which AI model leads HLE-Diamond (AA)?
Claude Opus 5.5 leads HLE-Diamond (AA) with 67.9% as of 28 Oct 2026, ahead of Claude Opus 5.5 at 64.0% (result dated 22 Sept 2026).
How many models have HLE-Diamond (AA) results?
132 model configurations have a HLE-Diamond (AA) result on BenchLeader, all taken from Artificial Analysis.
Who runs HLE-Diamond (AA) and how often is it updated?
HLE-Diamond (AA) is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does HLE-Diamond (AA) count toward the BenchLeader Index?
Yes. HLE-Diamond (AA) contributes to the reasoning category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.