Results: 119 models tested, 106 finished the full 40.

Top 15 of 106 full-coverage models

Mean score across all 40 tasks. Each model name opens its graded answers in the reviewer.

Scores compare within this run only. Frontier baselines carry a frontier tag. Modalities in means what you can send the model: text, images, files, video, or audio. Modalities out means what it can send back: text for 118 of 119 models, plus images for gemini-3-pro-image. Bench tasks feed text extracts, so modalities do not change the score. They are here so you know what each model can handle in real work.

All 106 full-coverage models

Ranked by mean score, best first. Model names link to graded answers in the reviewer.

13 partial runs, disclosed

These models stopped short of the full 40 tasks, mostly on slow responses that timed out. Ungraded tasks are excluded, never scored as zero, so these rows stay out of the ranking.

Every model, every task

The cell matrix shows one score per model per task: 106 ranked rows plus 13 partial rows across all 40 tasks, each cell linking to the graded answer.

Open the cell matrix

Take the data

Same run, two formats. Numbers match the boards above.