Benchmark-level results
Scores of 6 frontier general-purpose systems on each individual benchmark, grouped by the 9 areas of computer vision. The capability-level view on the project page aggregates these scores into 34 capabilities.
Higher is better unless marked ↓. The best score in each row is in bold, and shading is relative to that best score. Fractional scores are reported as percentages. n/s: the model does not support the required output type (image generation or editing).