TRAP - Image Task Accuracy: leaderboard

Metric: Task accuracy (%) on the 200 image-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 2029. 22 models tracked.

Top models

#ModelScore
1Qwen 3 VL 32B75.5
2GPT-5 Mini71.5
3GPT-5.4 Mini71
4Gemini 2.5 Flash67.5
5GPT-564.5
6GPT-4o Mini64
7Gemini 2.5 Pro62
8InternVL3.5-8B61
9Claude Sonnet 4.560
10Gemini 2.5 Flash Lite57
11Claude Haiku 4.555.5

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