pdfQA (Synthetic): leaderboard
Question answering over whole PDF documents (pdfQA; Schimanski et al., University of Zurich and ETH Zurich, 2026): syn-pdfQA holds GPT-4.1-generated questions over financial reports, research articles, books and sustainability disclosures that pass quality and difficulty filters, answered with the full PyMuPDF-parsed document in the prompt (documents under 128K tokens, 970 pairs). A gpt-4.1-mini G-Eval prompt scores each answer's correctness against the reference from 1 to 5; higher is better.
Metric: G-Eval correctness (1-5; gpt-4.1-mini judge; 970 syn-pdfQA pairs with documents under 128K tokens, full parsed document as context). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-OSS-120B | 4.4 | #699 |
| 2 | Qwen 3 Next 80B A3B | 4.06 | #756 |
| 3 | GPT-OSS-20B | 4.05 | #1150 |
| 4 | Gemma 3 27B | 3.73 | #1638 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pdfqa-synthetic · How It Works · Data refreshed daily, snapshot 2026-11-12.