Strongest agentic model at launch, sustained multi-hour autonomous coding
| Benchmark | Score | Rank |
|---|---|---|
ARC-C Grade-school science questions requiring reasoning | 97.2% | #16 / 40 |
ARC-AGIARC Prize Novel reasoning tasks requiring fluid intelligence | 8.6% | #19 / 23 |
HumanEval Coding ability - generating correct Python functions | 93.2% | #21 / 50 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 86.2% | #21 / 38 |
HellaSwag Common sense reasoning about everyday situations | 91.2% | #23 / 36 |
SWE-bench Real-world GitHub issue resolution | 72.5% | #23 / 40 |
MMLU Tests knowledge across 57 subjects from STEM to humanities | 88.8% | #26 / 54 |
MMMUvals.ai College-level multimodal reasoning across 30+ disciplines | 73.3% | #30 / 39 |
MATH Competition-level mathematics problems | 84.6% | #31 / 50 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 70.2% | #33 / 40 |
TerminalArtificial Analysis Agentic terminal coding tasks requiring multi-step execution | 31.1% | #33 / 48 |
Arena Elo Human preference ranking via blind comparisons | 1342 | #38 / 52 |
GPQA PhD-level science questions even experts struggle with | 79.6% | #40 / 64 |