BenchLeader

HLE Text-Only (Scale) leaderboard

As of 22 Nov 2026, the top score on HLE Text-Only (Scale) is 54.2% by GPT-6 Astra (OpenAI), ahead of Gemini 3.1 Pro at 47.3%, across 64 model configurations with a published result (Scale AI SEAL, result dated 9 Sep 2026). Humanity's Last Exam without its image questions, run by Scale AI: the only HLE score models without image input get. Not comparable with the full exam, which the BenchLeader Index uses. HLE Text-Only (Scale) is shown for reference only and does not count toward the BenchLeader Index.

Key facts (as of 22 Nov 2026)

  • GPT-6 Astra holds the top HLE Text-Only (Scale) score at 54.2% (Scale AI SEAL, 9 Sep 2026).
  • Gemini 3.1 Pro is second on HLE Text-Only (Scale) at 47.3% (Scale AI SEAL, 10 Apr 2026).
  • HLE Text-Only (Scale): Humanity's Last Exam without its image questions, run by Scale AI: the only HLE score models without image input get. Not comparable with the full exam, which the BenchLeader Index uses (Scale AI SEAL).
  • 64 model configurations have a published HLE Text-Only (Scale) result on BenchLeader (BenchLeader, 22 Nov 2026).
  • HLE Text-Only (Scale) does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).

Humanity's Last Exam without its image questions, run by Scale AI: the only HLE score models without image input get. Not comparable with the full exam, which the BenchLeader Index uses.

Scale AI SEAL marks 56 of these 64 results as possibly contaminated: the model was evaluated after the questions were public, so they may already be in its training data. Those scores are in the table below; this board does not feed the index.

Published by
Scale AI SEAL
Category
Reasoning
Index weight
Reference only
Models
64
Data as of
22 Nov 2026
Newest result
12 Oct 2026

Scale AI SEAL Leaderboards.

64 of 64
#
1GPT-6 Astra54.2%68.02026-09-09possible
2Gemini 3.1 Prohigh47.3%58.22026-04-10possible
3Claude Fable 5.1xhigh46.8%63.92026-09-03possible
4Claude Sonnet 5.5xhigh46.6%62.42026-10-12–
5GPT-5.4 Pro45.3%62.72026-03-23possible
6Gemini 3.8 Flash45.0%55.72026-09-09possible
7Muse Spark40.9%64.82026-04-08possible
8Gemini 3 Pro37.7%60.32025-11-19possible
9GPT-5.4thinking36.5%–2026-03-10possible
10Claude Opus 4.6max36.2%56.82026-02-17possible
11GPT-5 Pro33.3%61.32025-11-06possible
12Claude Haiku 5.530.4%–2026-10-08–
13GPT-5.228.5%55.42025-12-15possible
14Claude Opus 4.5thinking26.3%59.62025-11-26possible
15GPT-526.3%56.82025-08-07possible
16GPT-5.1thinking24.6%–2025-11-26possible
17Gemini 2.5 Pro22.1%53.52025-06-05possible
18o3high20.6%54.82025-04-16possible
19o3medium19.8%53.12025-04-16possible
20GPT-5 mini19.7%56.52025-08-22possible
21Claude Opus 4.6no reasoning19.4%57.12026-02-17possible
22o4-minihigh18.9%50.82025-04-16possible
23Gemini 2.5 Pro Preview 032518.4%59.72025-04-10possible
24Gemini 2.5 Pro Preview 050618.4%56.92025-05-06possible
25gpt-oss-120b15.5%46.22025-09-18possible
26Qwen3 235B A22B 2507thinking15.4%51.72025-08-05possible
27o4-minimedium14.5%51.42025-04-16possible
28Claude Sonnet 4.5thinking14.1%52.92025-10-02possible
29DeepSeek R1 052814.0%49.02025-06-02possible
30Claude Opus 4.513.9%56.92025-11-26possible
31o3-minihigh13.4%48.32025-04-10possible
32DeepSeek V3.112.9%48.52025-09-08possible
33Gemini 2.5 Flash12.6%52.32025-04-17possible
34Qwen3 235B A22B11.8%50.02025-05-01possible
35Claude Opus 4.1thinking11.3%58.02025-08-08possible
36Claude Opus 4thinking10.8%55.72025-05-24possible
37o3-minimedium10.3%46.22025-04-10possible
38gpt-oss-20b9.7%44.52025-09-18possible
39GLM-4.59.6%51.62025-08-13possible
40GLM-4.5-Air9.4%46.72025-08-13possible
41DeepSeek R18.5%47.02025-04-10–
42Gemini 3.1 Flash Lite8.0%52.52026-03-23possible
43Claude 3.7 Sonnetthinking7.9%52.52025-04-10possible
44o17.8%53.62025-04-10–
45o1-pro7.7%–2025-04-10possible
46Claude Sonnet 4.57.7%53.62025-10-02possible
47Claude Sonnet 4thinking7.6%53.82025-05-23possible
48Claude Opus 4.17.4%51.72025-08-08possible
49Gemini 2.0 Flash Thinking 01216.5%43.42025-04-10–
50GPT 5.1 Instant6.5%–2025-11-26possible
51Claude Opus 46.3%51.72025-05-23possible
52GPT-4.55.8%50.32025-04-10possible
53Claude Sonnet 45.4%49.92025-05-23possible
54Llama 4 Maverick5.3%41.72025-04-10possible
55GPT-4.15.0%48.52025-04-14possible
56Kimi K24.7%51.02025-07-23possible
57DeepSeek V34.5%44.22025-04-10possible
58Gemini 1.5 Pro 0024.5%43.72025-04-10–
59Nova Micro4.4%39.12025-04-10possible
60Mistral Medium 34.4%44.22025-05-13possible

AI reasoning and math benchmark leaderboard · All LLM benchmarks

Cite as: BenchLeader, “HLE Text-Only (Scale) leaderboard”, data as of 22 Nov 2026, https://www.benchleader.com/benchmarks/scale_hle_text_only.

HLE Text-Only (Scale): questions

What does HLE Text-Only (Scale) measure?
Humanity's Last Exam without its image questions, run by Scale AI: the only HLE score models without image input get. Not comparable with the full exam, which the BenchLeader Index uses. Scores are reported in percent of tasks solved; higher is better.
What is the top score on HLE Text-Only (Scale)?
As of 22 Nov 2026, the highest score on HLE Text-Only (Scale) is 54.2% by GPT-6 Astra (result dated 9 Sep 2026), ahead of Gemini 3.1 Pro at 47.3%. Scores come from Scale AI SEAL and are re-read daily.
Which AI model leads HLE Text-Only (Scale)?
GPT-6 Astra leads HLE Text-Only (Scale) with 54.2% as of 22 Nov 2026 (result dated 9 Sep 2026), ahead of Gemini 3.1 Pro at 47.3%.
How many models have HLE Text-Only (Scale) results?
64 model configurations have a HLE Text-Only (Scale) result on BenchLeader, all taken from Scale AI SEAL.
Who runs HLE Text-Only (Scale) and how often is it updated?
HLE Text-Only (Scale) is published by Scale AI SEAL. BenchLeader re-reads the published results once a day and records the date each result was published.
Does HLE Text-Only (Scale) count toward the BenchLeader Index?
No. HLE Text-Only (Scale) is shown for reference but left out of the composite index.