TRAP - Multimodal Privacy: leaderboard

Metric: Privacy score (%, 1 minus leakage) on the 200 multimodal-document scenarios of 500 document-grounded scenarios (100 text, 200 image, 200 multimodal) over ten private document types; each pairs a task query that needs a tool call with private fields and an attack query eliciting the same fields in natural language; rule-based evaluator, greedy decoding. Source: arxiv.org. Saturation forecast: Around 2035. 22 models tracked.

Top models

#ModelScore
1Gemini 2.5 Pro32
2Gemini 2.5 Flash Lite18
3GPT-4o Mini13.5
4InternVL3.5-8B12.5
5GPT-511.5
6Gemini 2.5 Flash9
7GPT-5 Mini8
8GPT-5.4 Mini8
9Qwen 3 VL 32B8
10Claude Haiku 4.56
11Claude Sonnet 4.53.5

Interactive version: theaggregate.ai/benchmark?slug=trap-multimodal-privacy · How It Works · Data refreshed daily, snapshot 2026-09-29.