First model to break 80% on SWE-bench Verified
| Benchmark | Score | Rank |
|---|---|---|
SWE-bench Real-world GitHub issue resolution | 80.9% | #2 / 38 |
HumanEval Coding ability - generating correct Python functions | 96.4% | #3 / 49 |
MATH Competition-level mathematics problems | 100% | #3 / 49 |
ARC-C Grade-school science questions requiring reasoning | 98.6% | #3 / 40 |
MMLU Tests knowledge across 57 subjects from STEM to humanities | 92.8% | #5 / 53 |
Terminal Agentic terminal coding tasks requiring multi-step execution | 59.8% | #6 / 37 |
HellaSwag Common sense reasoning about everyday situations | 96.8% | #7 / 36 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 87.3% | #9 / 30 |
MMMUvals.ai College-level multimodal reasoning across 30+ disciplines | 82.9% | #9 / 33 |
ARC-AGI Novel reasoning tasks requiring fluid intelligence | 37.6% | #10 / 21 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 83.7% | #12 / 31 |
Arena Elo Human preference ranking via blind comparisons | 1445 | #14 / 41 |
GPQA PhD-level science questions even experts struggle with | 87% | #18 / 54 |