SynthDocBench: leaderboard
SynthDocBench (ServiceNow, 2026) is a synthetic long-document visual question answering benchmark: 1,788 questions over 200 generated reports averaging 51 pages, read as rasterized page images. The board is accuracy in percent over all questions, where a GPT-5 judge scores each answer against a deterministic reference and a score of 6 of 10 or more counts as correct; higher is better.
Metric: Accuracy (%) over all 1,788 questions: GPT-5 judge scores each answer 0 to 10 against a deterministic reference and an answer counts when the score is at least 6; pages rasterized at 144 DPI in 5-page strips, vision-only input, temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 72.5 | #94 |
| 2 | Qwen 3 VL 235B A22B | 58.6 | #479 |
| 3 | GPT-5.4 | 42.3 | #123 |
| 4 | GPT-4o | 38.6 | #741 |
| 5 | InternVL3-78B | 38.3 | #738 |
| 6 | Claude Sonnet 4.5 | 31.4 | #250 |
| 7 | Qwen 2.5 VL 7B Instruct | 8.1 | #1833 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=synthdocbench · How It Works · Data refreshed daily, snapshot 2026-11-13.