Humanity's Last Exam (Epoch AI replication)

Humanity's Last Exam (Epoch AI replication) is run by Epoch AI (AI Benchmarking Hub). It has ranked 51 models, of which the catalogue holds 37, scoring from 2.72 to 46.5 on Accuracy.

measured by Epoch AI (AI Benchmarking Hub) · the board itself

Place	Model	Metric	Score
1stof 51	Claude Fable 5.1	Accuracy (xhigh)	46.5
2ndof 51	Gemini 3.1 Pro Preview	Accuracy	46.44
3rdof 51	GPT-5.4 Pro	Accuracy	44.32
4thof 51	Muse Spark 1.2	Accuracy	40.56
5thof 51	gemini-3-pro-preview	Accuracy	37.52
6thof 51	GPT-5.4	Accuracy (xhigh)	36.24
7thof 51	Claude Opus 4.7	Accuracy	36.2
8thof 51	Claude Opus 4.6	Accuracy (max)	34.44
9thof 51	GPT-5 Pro	Accuracy	31.64
10thof 51	GPT-5.2	Accuracy	27.8
11thof 51	GPT-5	Accuracy (high)	25.32
12thof 51	Claude Opus 4.5	Accuracy	25.2
13thof 51	Kimi K2.5	Accuracy	24.37
14thof 51	GPT-5.1	Accuracy	23.68
15thof 51	Gemini 2.5 Pro Preview 06-05	Accuracy	21.64
16thof 51	o3	Accuracy (high)	20.32
17thof 51	GPT-5 mini	Accuracy	19.44
21stof 51	o4-mini	Accuracy (high)	18.08
22ndof 51	Gemini 2.5 Pro Preview 05-06	Accuracy	17.8
25thof 51	Claude Sonnet 4.5	Accuracy	13.72
26thof 51	Gemini 2.5 Flash	Accuracy	12.08
27thof 51	Claude Opus 4.1	Accuracy	11.52
29thof 51	Claude Opus 4	Accuracy	10.72
30thof 51	Gemini 3.1 Flash-Lite	Accuracy	8.64
31stof 51	GLM-4.5	Accuracy	8.32
32ndof 51	GLM-4.5-Air	Accuracy	8.12
33rdof 51	o1-pro	Accuracy	8.12
34thof 51	Claude 3.7 Sonnet	Accuracy	8.04
35thof 51	o1	Accuracy	7.96
37thof 51	Claude Sonnet 4	Accuracy	7.76
42ndof 51	Llama-4-Maverick-17B-128E-Instruct	Accuracy	5.68
45thof 51	GPT-4.1 fine-tuned	Accuracy	5.4
47thof 51	Mistral Medium 2505	Accuracy	4.52
48thof 51	Nova Pro	Accuracy	4.4
49thof 51	Claude 3.5 Sonnet	Accuracy	4.08
50thof 51	Nova Lite	Accuracy	3.64
51stof 51	GPT-4o (2024-11-20)	Accuracy	2.72

Place

Model

Metric

Score

1stof 51

Claude Fable 5.1

Accuracy (xhigh)

46.5

2ndof 51

Gemini 3.1 Pro Preview

Accuracy

46.44