First hybrid reasoning model, extended thinking mode for complex problems
| Benchmark | Score | Rank |
|---|---|---|
MATH Competition-level mathematics problems | 96.2% | #14 / 50 |
HumanEval Coding ability - generating correct Python functions | 93.7% | #18 / 50 |
ARC-C Grade-school science questions requiring reasoning | 96.7% | #24 / 40 |
MMMUvals.ai College-level multimodal reasoning across 30+ disciplines | 75.1% | #25 / 39 |
HellaSwag Common sense reasoning about everyday situations | 89% | #28 / 36 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 82.7% | #29 / 38 |
SWE-bench Real-world GitHub issue resolution | 62.3% | #30 / 40 |
GPQA PhD-level science questions even experts struggle with | 84.8% | #31 / 64 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 60.4% | #36 / 40 |
TerminalArtificial Analysis Agentic terminal coding tasks requiring multi-step execution | 21.2% | #39 / 48 |
MMLU Tests knowledge across 57 subjects from STEM to humanities | 86.1% | #41 / 54 |
Arena Elo Human preference ranking via blind comparisons | 1310 | #41 / 52 |