Humanities Last Exam
An AI evaluation benchmark spanning dozens of obscure and complex humanities disciplines, designed to assess model reasoning on high-difficulty academic questions in the humanities.
Timeline (last 90 days)
DeepSeek V4.1 Flash scored 42 on Humanities Last Exam, tied with Kimi K3 and GLM 5.3; its Terminal Bench score was also 42.
Humanities Last Exam is an expert-level question bank covering humanities disciplines such as philosophy, history, and literature, where top models generally have a correct answer rate below 20%.
On the Diamond version of Humanities Last Exam, Astra surpasses Opus 5.5 by approximately 5%
Claude 5.1 scored 60.9% on Humanities Last Exam without tools, rising to 65% with tools enabled.
All Facts (3)
DeepSeek V4.1 Flash scored 42 on Humanities Last Exam, tied with Kimi K3 and GLM 5.3; its Terminal Bench score was also 42.
50%UnverifiedHumanities Last Exam is an expert-level question bank covering humanities disciplines such as philosophy, history, and literature, where top models generally have a correct answer rate below 20%.
50%UnverifiedClaude 5.1 scored 60.9% on Humanities Last Exam without tools, rising to 65% with tools enabled.
50%