Flash-tier model outperforming previous-gen Pro on most benchmarks
| Benchmark | Score | Rank |
|---|---|---|
ARC-C Grade-school science questions requiring reasoning | 98.4% | #6 / 40 |
HellaSwag Common sense reasoning about everyday situations | 97.1% | #6 / 36 |
MMMUvals.ai College-level multimodal reasoning across 30+ disciplines | 87.6% | #9 / 53 |
HumanEval Coding ability - generating correct Python functions | 94.8% | #12 / 50 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 88.6% | #14 / 52 |
MMLU Tests knowledge across 57 subjects from STEM to humanities | 91.2% | #16 / 54 |
ARC-AGIARC Prize Novel reasoning tasks requiring fluid intelligence | 34% | #18 / 28 |
MATH Competition-level mathematics problems | 94.6% | #19 / 50 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 85.6% | #19 / 54 |
GPQA PhD-level science questions even experts struggle with | 90.4% | #24 / 81 |
SWE-bench Real-world GitHub issue resolution | 72% | #24 / 39 |
Arena Elo Human preference ranking via blind comparisons | 1428 | #24 / 51 |
hleArtificial Analysis | 34.7% | #28 / 69 |
TerminalArtificial Analysis Agentic terminal coding tasks requiring multi-step execution | 38.6% | #33 / 56 |