Skip to content
Weights Not Walls
Menu

HLE

Reported

Humanity's Last Exam: hard questions written by experts across many academic subjects, many with a single short answer.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Score shown
Epoch's column "Accuracy", as a percentage, higher is better. Error bars as the source gives them.
Results
56 in all: 4 for open-weight models, 52 for closed ones. 4 linked to a model page here.

Best open-weight result: moonshotai/Kimi-K2.5, 24.4%, number 18 of 56 results.

HLE results
#ModelWeightsSettingScoreDateSource
1gpt-6-astra_unknownOpenAIclosedeffort unknown54.8%±1.9not datedvia Epoch AI
2claude-fable-5-1_xhighAnthropicclosedeffort xhigh46.5%±2.0not datedvia Epoch AI
3gemini-3.1-pro-previewGoogle DeepMindcloseddefault46.4%±2.0not datedvia Epoch AI
4claude-sonnet-5-5_xhighAnthropicclosedeffort xhigh46.1%±1.9not datedvia Epoch AI
5gemini-3.8-flash_unknownGoogle DeepMindclosedeffort unknown44.5%±2.0not datedvia Epoch AI
6gpt-5.4-pro-2026-03-05_unknownOpenAIclosedeffort unknown44.3%±1.9not datedvia Epoch AI
7muse-sparkMeta AIcloseddefault40.6%±1.9not datedvia Epoch AI
8gemini-3-pro-previewGoogle DeepMindcloseddefault37.5%±1.9not datedvia Epoch AI
9gpt-5.4-2026-03-05_xhighOpenAIclosedeffort xhigh36.2%±1.9not datedvia Epoch AI
10claude-opus-4-7_unknownAnthropicclosedeffort unknown36.2%±1.9not datedvia Epoch AI
11claude-opus-4-6_maxAnthropicclosedeffort max34.4%±1.9not datedvia Epoch AI
12gpt-5-pro-2025-10-06_unknownOpenAIclosedeffort unknown31.6%±1.8not datedvia Epoch AI
13claude-haiku-5-5_unknownAnthropicclosedeffort unknown30.1%±1.8not datedvia Epoch AI
14gpt-5.2-2025-12-11_unknownOpenAIclosedeffort unknown27.8%±1.8not datedvia Epoch AI
15gpt-5-2025-08-07_highOpenAIclosedeffort high25.3%±1.7not datedvia Epoch AI
16gpt-5-2025-08-07_unknownOpenAIclosedeffort unknown25.3%±1.7not datedvia Epoch AI
17claude-opus-4-5-20251101_unknownAnthropicclosedeffort unknown25.2%±1.7not datedvia Epoch AI
18moonshotai/Kimi-K2.5Epoch: kimi-k2.5, Moonshotopendefault24.4%±1.8not datedvia Epoch AI
19gpt-5.1-2025-11-13_unknownOpenAIclosedeffort unknown23.7%±1.7not datedvia Epoch AI
20gemini-2.5-pro-preview-06-05Google DeepMindcloseddefault21.6%±1.6not datedvia Epoch AI
21o3-2025-04-16_highOpenAIclosedeffort high20.3%±1.6not datedvia Epoch AI
22gpt-5-mini-2025-08-07_unknownOpenAIclosedeffort unknown19.4%±1.6not datedvia Epoch AI
23o3-2025-04-16_mediumOpenAIclosedeffort medium19.2%±1.5not datedvia Epoch AI
24claude-opus-4-6Anthropiccloseddefault19.0%±1.5not datedvia Epoch AI
25gemini-2.5-pro-exp-03-25Google DeepMindcloseddefault18.2%±1.5not datedvia Epoch AI
26o4-mini-2025-04-16_highOpenAIclosedeffort high18.1%±1.5not datedvia Epoch AI
27gemini-2.5-pro-preview-05-06Google DeepMindcloseddefault17.8%±1.5not datedvia Epoch AI
28o4-mini-2025-04-16_mediumOpenAIclosedeffort medium14.3%±1.4not datedvia Epoch AI
29claude-opus-4-5-20251101_unknownAnthropicclosedeffort unknown14.2%±1.4not datedvia Epoch AI
30claude-sonnet-4-5-20250929_unknownAnthropicclosedeffort unknown13.7%±1.4not datedvia Epoch AI
31gemini-2.5-flash-preview-04-17Google DeepMindcloseddefault12.1%±1.3not datedvia Epoch AI
32claude-opus-4-1-20250805_unknownAnthropicclosedeffort unknown11.5%±1.3not datedvia Epoch AI
33gemini-2.5-flash-preview-05-20Google DeepMindcloseddefault11.0%±1.2not datedvia Epoch AI
34claude-opus-4-20250514_unknownAnthropicclosedeffort unknown10.7%±1.2not datedvia Epoch AI
35gemini-3.1-flash-liteGooglecloseddefault8.6%±1.1not datedvia Epoch AI
36zai-org/GLM-4.5Epoch: glm-4.5, Z.ai (Zhipu AI),Tsinghua Universityopendefault8.3%±1.1not datedvia Epoch AI
37zai-org/GLM-4.5-AirEpoch: GLM-4.5-Air, Z.ai (Zhipu AI),Tsinghua Universityopendefault8.1%±1.1not datedvia Epoch AI
38o1-pro-2025-03-19OpenAIcloseddefault8.1%±1.1not datedvia Epoch AI
39claude-3-7-sonnet-20250219_unknownAnthropicclosedeffort unknown8.0%±1.1not datedvia Epoch AI
40o1-2024-12-17_unknownOpenAIclosedeffort unknown8.0%±1.1not datedvia Epoch AI
41claude-opus-4-1-20250805_unknownAnthropicclosedeffort unknown7.9%±1.1not datedvia Epoch AI
42claude-sonnet-4-20250514_unknownAnthropicclosedeffort unknown7.8%±1.1not datedvia Epoch AI
43claude-sonnet-4-5-20250929_unknownAnthropicclosedeffort unknown7.5%±1.0not datedvia Epoch AI
44gpt-5.1-2025-11-13_noneOpenAIclosedeffort none6.8%±1.0not datedvia Epoch AI
45claude-opus-4-20250514_unknownAnthropicclosedeffort unknown6.7%±1.0not datedvia Epoch AI
46gemini-2.0-flash-thinking-exp-01-21Google DeepMind,Googlecloseddefault6.6%±1.0not datedvia Epoch AI
47meta-llama/Llama-4-Maverick-17B-128E-InstructEpoch: Llama-4-Maverick-17B-128E-Instruct, Meta AIopendefault5.7%±0.9not datedvia Epoch AI
48claude-sonnet-4-20250514_unknownAnthropicclosedeffort unknown5.5%±0.9not datedvia Epoch AI
49gpt-4.5-preview-2025-02-27OpenAIcloseddefault5.4%±0.9not datedvia Epoch AI
50gpt-4.1-2025-04-14OpenAIcloseddefault5.4%±0.9not datedvia Epoch AI
51gemini-1.5-pro-002Google DeepMindcloseddefault4.6%±0.8not datedvia Epoch AI
52mistral-medium-2505Mistral AIcloseddefault4.5%±0.8not datedvia Epoch AI
53amazon.nova-pro-v1:0Amazoncloseddefault4.4%±0.8not datedvia Epoch AI
54claude-3-5-sonnet-20241022Anthropiccloseddefault4.1%±0.8not datedvia Epoch AI
55amazon.nova-lite-v1:0Amazoncloseddefault3.6%±0.7not datedvia Epoch AI
56gpt-4o-2024-11-20OpenAIcloseddefault2.7%±0.6not datedvia Epoch AI

Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method