VeriOCRBench - Factual Dimension (Assisted): leaderboard
The factual verification dimension of VeriOCRBench (Zhou et al., 2026): 400 OCR-grounded tasks with the traps fact tampering and vague reference, where a stated attribute or value contradicts the image or a reference has several plausible targets. Diagnosis accuracy is the percentage on which the model both rejects the task and names the correct trap mechanism, judged by GPT-4o; higher is better. Assisted setting: the prompt tells the model to check that the task is completable before answering.
Metric: Diagnosis accuracy (%; the 400 trap-injected tasks of the factual verification dimension (fact tampering and vague reference: a stated attribute or value contradicts the visible evidence, or a reference has several plausible targets); the model must reject the task and name the ground-truth trap mechanism; judged by GPT-4o at temperature 0; assisted paradigm: a meta-instruction asks the model to verify the task first). Source: arxiv.org. Saturation forecast: Around 2028. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 397B A17B | 78 | #242 |
| 2 | GPT-5 | 73 | #163 |
| 3 | Claude Sonnet 4.5 | 67.75 | #244 |
| 4 | Qwen 3.5 35B A3B | 64.75 | #480 |
| 5 | Qwen 3.5 122B A10B | 63.5 | #329 |
| 6 | Llama 4 Maverick | 56 | #1105 |
| 7 | GPT-4o | 47.25 | #741 |
| 8 | InternVL3-38B | 45 | #905 |
| 9 | Llama 4 Scout | 37.25 | #1787 |
| 10 | Gemini 3.1 Pro (Preview) | 7 | #91 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=veriocrbench-factual-dimension-assisted · How It Works · Data refreshed daily, snapshot 2026-11-12.