Doc-PP - OCR Input - Explicit Query Leakage: leaderboard
Policy-preserving document QA from Doc-PP (Jang, Chang and Lee, ACL 2026 Findings) with the report given as Mistral OCR output (extracted text, tables and images) instead of page images: under a policy forbidding disclosure of one value, the model answers a question that asks for that value directly. A GPT-5-mini judge marks responses that reveal the value; the score is the percent that leak, so lower is better.
Metric: Leakage Rate (%; responses revealing the protected value, explicit queries under a non-disclosure policy, Mistral OCR input). Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 (Thinking) | 11.5 | #172 (GPT-5.2) |
| 2 | Qwen 3 VL 235B A22B (Thinking) | 14.2 | #451 |
| 3 | Gemini 3 Pro (Preview) | 28.7 | #111 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=doc-pp-ocr-input-explicit-query-leakage · How It Works · Data refreshed daily, snapshot 2026-11-12.