1M context, stronger coding and vision. SWE-bench Pro 64.3% beats GPT-5.4 and Gemini 3.1 Pro.
| Benchmark | Score | Rank |
|---|---|---|
SWE-bench Real-world GitHub issue resolution | 87.6% | #1 / 39 |
Arena Elo Human preference ranking via blind comparisons | 1753 | #1 / 51 |
GPQA PhD-level science questions even experts struggle with | 94.2% | #2 / 73 |
OSWorld Computer use in real desktop environments | 78% | #2 / 13 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 89.9% | #4 / 45 |
ARC-AGIARC Prize Novel reasoning tasks requiring fluid intelligence | 82.2% | #4 / 26 |
Terminal Agentic terminal coding tasks requiring multi-step execution | 69.4% | #10 / 56 |
MMMUvals.ai College-level multimodal reasoning across 30+ disciplines | 85.5% | #11 / 46 |
MMLU Tests knowledge across 57 subjects from STEM to humanities | 91.5% | #14 / 54 |
hleArtificial Analysis | 39.6% | #14 / 61 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 84.7% | #20 / 48 |