Introducing the next generation of Claude

Benchmark Model Rank Results
arithmetic-reasoning-on-gsm8kClaude 3 Haiku–Accuracy: 88.9
arithmetic-reasoning-on-gsm8kClaude 3 Opus–Accuracy: 95.0
arithmetic-reasoning-on-gsm8kClaude 3 Sonnet–Accuracy: 92.3
code-generation-on-humanevalClaude 3 Opus–Pass@1: 84.9
math-word-problem-solving-on-mathClaude 3 Haiku–Accuracy: 38.9
math-word-problem-solving-on-mathClaude 3 Opus–Accuracy: 60.1
math-word-problem-solving-on-mathClaude 3 Sonnet–Accuracy: 43.1
math-word-problem-solving-on-mgsm-0-shotClaude 3 Haiku–Accuracy: 75.1
math-word-problem-solving-on-mgsm-0-shotClaude 3 Opus–Accuracy: 90.7
math-word-problem-solving-on-mgsm-0-shotClaude 3 Sonnet–Accuracy: 83.5
multi-task-language-understanding-on-mmluClaude 3 Haiku–Average (%): 75.2
multi-task-language-understanding-on-mmluClaude 3 Opus–Average (%): 86.8
multi-task-language-understanding-on-mmluClaude 3 Sonnet–Average (%): 79.0
multimodal-reasoning-on-mmmuClaude 3 Haiku–Accuracy: 50.2
multimodal-reasoning-on-mmmuClaude 3 Opus–Accuracy: 59.4
multimodal-reasoning-on-mmmuClaude 3 Sonnet–Accuracy: 53.1
multiple-choice-question-answering-mcqa-on-gpqa-diamondClaude 3 Haiku–Accuracy: 33.3
multiple-choice-question-answering-mcqa-on-gpqa-diamondClaude 3 Opus–Accuracy: 50.4
multiple-choice-question-answering-mcqa-on-gpqa-diamondClaude 3 Sonnet–Accuracy: 40.4