PPT-Eval - Average Score: leaderboard
PPT-Eval (Microsoft, 2026) evaluates computer-use agents on 120 real PowerPoint creation and editing tasks in PowerPoint for the web, where the agent acts through GUI actions from full-screen screenshots. The score is the average rubric score in percent: task-specific rubrics judged by Claude-4-Sonnet give partial credit for intermediate steps and penalize unnecessary changes, averaged over three runs. Higher is better.
Metric: Average rubric score (%) over all 120 PPT-Eval tasks: task-specific rubrics award partial credit for intermediate steps and penalize unnecessary changes and poor aesthetics, the agent editing the deck in PowerPoint for the web through GUI actions from full-screen screenshots; rubrics judged with Claude-4-Sonnet; mean of three runs; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 57 | #156 |
| 2 | Claude Sonnet 4 | 53 | #400 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ppt-eval-average-score · How It Works · Data refreshed daily, snapshot 2026-11-13.