Introducing Claude Opus 5

Benchmark Model Rank Results
agents-on-toolathlon-verified-108-tasksClaude Opus 5#1Pass@1: 0.806
code-generation-on-swe-bench-verifiedClaude Opus 5#1Accuracy: 96.0
computer-use-agents-on-osworld-2-0Claude Opus 5 (Anthropic system card)#1Success Rate: 70.57
multi-task-language-understanding-on-includeClaude Opus 5#1Accuracy: 89.8
code-generation-on-swe-bench-proClaude Opus 5#3Resolved: 79.2
exploit-generation-on-exploitbenchClaude Opus 5 (ExploitBench AutoNudge)#3Coverage: 0.7
code-generation-on-swe-bench-multilingualClaude Opus 5#4Accuracy: 89.5
medical-question-answering-on-healthbench-professionalClaude Opus 5#4Score: 59.8
abstract-reasoning-on-arc-agi-2Claude Opus 5#5Accuracy: 90.42
abstract-reasoning-on-arc-agiClaude Opus 5#7Accuracy: 97.5
code-generation-on-deepsweClaude Opus 5#11Pass@1: 0.688