Open-weight 397B MoE with visual agentic capabilities, 201 languages
| Benchmark | Score | Rank |
|---|---|---|
OSWorld Computer use in real desktop environments | 62.2% | #8 / 13 |
SWE-bench Real-world GitHub issue resolution | 76.4% | #12 / 39 |
Terminal Agentic terminal coding tasks requiring multi-step execution | 52.5% | #16 / 56 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 85.3% | #18 / 48 |
MMMUvals.ai College-level multimodal reasoning across 30+ disciplines | 81.9% | #19 / 46 |
Arena Elo Human preference ranking via blind comparisons | 1451 | #21 / 51 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 87.2% | #22 / 45 |
GPQA PhD-level science questions even experts struggle with | 88.4% | #27 / 73 |
MMLU Tests knowledge across 57 subjects from STEM to humanities | 88.5% | #31 / 54 |
HumanEval Coding ability - generating correct Python functions | 84.8% | #39 / 50 |
hleArtificial Analysis | 13.9% | #45 / 61 |