Humanity's Last Exam
An extremely difficult knowledge-reasoning question bank constructed through academic crowdsourcing, covering frontier problems in mathematics, science, the humanities, and other domains. It is designed to evaluate the knowledge-reasoning limits of top-tier AI models by probing the boundaries of current human and machine understanding.
Timeline (last 90 days)
Humanity's Last Exam is an advanced reasoning benchmark jointly launched by Scale AI and the Center for AI Safety, released in early 2025, containing about 3,000 expert-level questions.
GPQA Diamond, MMLU-Pro, Humanity's Last Exam, SWE-bench Verified, Terminal-Bench, MMMU-Pro, LEXam, and Open ASR Leaderboard are all official benchmarks marked by Hugging Face as official benchmark.
Opus leads on Terminal-Bench 4.0, Humanity's Last Exam and long-context reasoning.
Mythos scored 64.7% on Humanity's Last Exam, far exceeding any currently publicly available model.
Haiku 5.5's score on Humanity's Last Exam improved from 10.2% to 45.9% without tools, and from 18.7% to 57.4% with tools.
Humanity's Last Exam was designed by hundreds of academic experts, covering extremely difficult questions in mathematics, science, humanities, and other fields, and is used to test the knowledge boundaries and reasoning depth of models.
Opus 5.5 scored 67 on Humanity's Last Exam, Astra scored 57, and Fable 5.1 scored 65.
On Terminal Bench 3.0, Opus 5 leads with 43.3 versus V4.1's 30.0, and on Humanity's Last Exam, Opus 5 scores 56.3 while V4.1 only reaches 36.8.
DeepSeek 4.1 Flash can surpass top-tier models such as Claude Opus 5 and Kimi K3 on certain benchmarks, but does not lead across the board.
Humanity's Last Exam由Scale AI与全球数百位顶尖学者联合推出，旨在创建一个不会很快被饱和的极难评测基准
All Facts (9)
Humanity's Last Exam is an advanced reasoning benchmark jointly launched by Scale AI and the Center for AI Safety, released in early 2025, containing about 3,000 expert-level questions.
50%UnverifiedGPQA Diamond, MMLU-Pro, Humanity's Last Exam, SWE-bench Verified, Terminal-Bench, MMMU-Pro, LEXam, and Open ASR Leaderboard are all official benchmarks marked by Hugging Face as official benchmark.
50%UnverifiedOpus leads on Terminal-Bench 4.0, Humanity's Last Exam and long-context reasoning.
50%UnverifiedMythos scored 64.7% on Humanity's Last Exam, far exceeding any currently publicly available model.
50%UnverifiedHaiku 5.5's score on Humanity's Last Exam improved from 10.2% to 45.9% without tools, and from 18.7% to 57.4% with tools.
50%UnverifiedHumanity's Last Exam was designed by hundreds of academic experts, covering extremely difficult questions in mathematics, science, humanities, and other fields, and is used to test the knowledge boundaries and reasoning depth of models.
50%UnverifiedOpus 5.5 scored 67 on Humanity's Last Exam, Astra scored 57, and Fable 5.1 scored 65.
50%UnverifiedOn Terminal Bench 3.0, Opus 5 leads with 43.3 versus V4.1's 30.0, and on Humanity's Last Exam, Opus 5 scores 56.3 while V4.1 only reaches 36.8.
50%UnverifiedHumanity's Last Exam由Scale AI与全球数百位顶尖学者联合推出，旨在创建一个不会很快被饱和的极难评测基准
50%