Opus-level performance at one-fifth the cost, default model on claude.ai
| Benchmark | Score | Rank |
|---|---|---|
OSWorld Computer use in real desktop environments | 72.5% | #7 / 9 |
SWE-bench Real-world GitHub issue resolution | 79.6% | #8 / 40 |
ARC-AGI Novel reasoning tasks requiring fluid intelligence | 60.4% | #8 / 23 |
Terminal Agentic terminal coding tasks requiring multi-step execution | 59.1% | #10 / 48 |
MMMUvals.ai College-level multimodal reasoning across 30+ disciplines | 83.6% | #11 / 39 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 87.3% | #15 / 38 |
Arena Elo Human preference ranking via blind comparisons | 1459 | #19 / 52 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 82.1% | #21 / 40 |
MMLU Tests knowledge across 57 subjects from STEM to humanities | 89.3% | #24 / 54 |
HumanEval Coding ability - generating correct Python functions | 92.1% | #27 / 50 |
MATH Competition-level mathematics problems | 85.3% | #30 / 50 |
GPQA PhD-level science questions even experts struggle with | 74.1% | #45 / 64 |