Humanity's Last Exam

Humanity's Last Exam is run by Center for AI Safety + Scale AI (agi.safe.ai / lastexam.ai). It has ranked 10 models, of which the catalogue holds 10, scoring from 2.7 to 38.3 on Accuracy (%).

measured by Center for AI Safety + Scale AI (agi.safe.ai / lastexam.ai) · the board itself

Place	Model	Metric	Score
1stof 10	gemini-3-pro-preview	Accuracy (%)	38.3
2ndof 10	GPT-5	Accuracy (%)	25.3
3rdof 10	Grok 4	Accuracy (%)	24.5
4thof 10	Gemini 2.5 Pro	Accuracy (%)	21.6
5thof 10	GPT-5 mini	Accuracy (%)	19.4
6thof 10	Claude Sonnet 4.5	Accuracy (%)	13.7
7thof 10	Gemini 2.5 Flash	Accuracy (%)	12.1
8thof 10	DeepSeek-R1	Accuracy (%)	8.5
9thof 10	o1	Accuracy (%)	8
10thof 10	GPT-4o	Accuracy (%)	2.7

Place

Model

Metric

Score

1stof 10

gemini-3-pro-preview

Accuracy (%)

38.3

2ndof 10

GPT-5

Accuracy (%)

25.3