SpatialTrust - Direct Factor Identification: leaderboard
SpatialTrust (University of Washington, 2026) tests whether multimodal LLMs recognize, localize and explain environmental information-leakage risks in laptop-camera authentication scenes, with evidence grounding at an IoU threshold of 0.5. Score for grounded identification of direct sensitive factors that immediately disclose information, in percent, higher is better.
Metric: Grounded identification score (%; direct sensitive factors that immediately disclose information; over SpatialTrust's 577 secure-authentication images (five questions each); IoU threshold 0.5 for grounded evidence). Source: arxiv.org. Saturation forecast: Around March 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 40.81 | #452 |
| 2 | Qwen 3 VL 30B A3B Instruct | 39.17 | #870 |
| 3 | Qwen 3 VL 8B Instruct | 28.68 | #982 |
| 4 | GPT-5.4 | 26.08 | #121 |
| 5 | GLM-4.1V-9B (Thinking) | 14.21 | #1061 (GLM-4.1V-9B) |
| 6 | Qwen 2 VL 7B Instruct | 6.76 | #2569 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=spatialtrust-direct-factor-identification · How It Works · Data refreshed daily, snapshot 2026-11-12.