PPTBench: leaderboard
PPTBench (Wang et al., 2026) asks coding agents to rebuild a scientific flow diagram, given only its image, as one natively editable PowerPoint slide, over 500 frozen tasks drawn from arXiv papers. An agentic judge applies validity, semantic and rendering gates and deducts for layout, text and local-graphics defects, giving a mean score from 0 to 100 with each run keeping its agent harness and reasoning effort; higher is better.
Metric: Mean reconstruction score (0-100): per task, the product of the artifact-validity, semantic-correctness and rendering-quality gates (majority of three judge rounds) and the layout (max 30), text (max 40) and local-graphics (max 30) scores left after severity-weighted defect deductions; invalid artifacts and gate failures score 0; 500 frozen tasks from arXiv flow diagrams, one run each; Agentic Judge GPT-5.6 Luna High, three rounds). Source: arxiv.org. Saturation forecast: Around August 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K3 (High) | 67.8 | #53 (Kimi K3) |
| 2 | GPT-5.6 Sol (Max) | 49.28 | #29 (GPT-5.6 Sol) |
| 3 | Qwen 3.8 Max (xHigh) | 47.69 | #55 (Qwen 3.8 Max) |
| 4 | GPT-5.6 Sol (xHigh) | 36.24 | #29 (GPT-5.6 Sol) |
| 5 | Claude Opus 5 (xHigh) | 35.91 | #20 (Claude Opus 5) |
| 6 | Claude Opus 5 (Max) | 33.54 | #20 (Claude Opus 5) |
| 7 | GPT-5.6 Terra (Max) | 33.46 | #60 (GPT-5.6 Terra) |
| 8 | Claude Opus 5 (High) | 33.12 | #20 (Claude Opus 5) |
| 9 | GLM-5.3 Flash (Max) | 32.42 | #119 (GLM-5.3 Flash) |
| 10 | GPT-5.6 Luna (Max) | 31.02 | #117 (GPT-5.6 Luna) |
| 11 | GPT-5.6 Sol (High) | 28.28 | #29 (GPT-5.6 Sol) |
| 12 | GPT-5.6 Terra (xHigh) | 25.84 | #60 (GPT-5.6 Terra) |
| 13 | GPT-5.6 Luna (xHigh) | 24.63 | #117 (GPT-5.6 Luna) |
| 14 | Claude Opus 5 (Medium) | 24.13 | #20 (Claude Opus 5) |
| 15 | GPT-5.6 Sol (Medium) | 22.2 | #29 (GPT-5.6 Sol) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pptbench · How It Works · Data refreshed daily, snapshot 2026-11-12.