Caractéristique | GPT-4.5 | GPT-4o |
---|---|---|
Indicateurs de performance | ||
GPQA (science) | 71.4% | 53.6% |
AIME '24 (mathématiques) | 36.7% | 9.3% |
MMMLU (multilingue) | 85.1% | 81.5% |
MMMU (multimodal) | 74.4% | 69.1% |
SWE-lancer Diamond (codage) | 32.6% | 23.3% |
SWE-Bench Verified (codage) | 38.0% | 30.7% |