Open-weight multimodal MoE trained jointly on text, images, audio, and video.
| Benchmark | Score | Rank |
|---|---|---|
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 85.5% | #21 / 54 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 86.3% | #31 / 52 |
MMMUvals.ai College-level multimodal reasoning across 30+ disciplines | 78.5% | #33 / 53 |
hleArtificial Analysis | 29.7% | #33 / 69 |
GPQAArtificial Analysis PhD-level science questions even experts struggle with | 87.2% | #42 / 81 |