Classement des modèles IA
Comparez les modèles IA sur les benchmarks agentiques, de raisonnement, de codage et d'utilisation d'outils.
Benchmarks
agentic
reasoning
coding
tool-use
computer-use
Humanity's Last Exam
2500+ expert-level questions across 100+ fields (math, science, humanities), crowd-sourced from domain experts. The hardest static benchmark — designed to be unsolvable by current AI, with top models scoring ~75%.
Reasoning19 modèles · % accuracy
