SafeRelBench - Proximity: leaderboard

Metric: Safety success rate (%) on proximity-relation samples: episodes that both complete the BDDL task goal and satisfy every process-level and termination-level safety condition, over all samples; closed-loop step-level planning in Behavior-1K household scenes, base prompt, deterministic decoding; higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 7 models tracked.

Top models

#ModelScore
1Claude Sonnet 4.638
2GPT-5.437
3Qwen 3 VL 32B Instruct35
4Qwen 2.5 VL 72B Instruct29
5Gemini 3.1 Pro (Preview)21
6Qwen 2.5 VL 32B Instruct21
7Qwen 3 VL 8B Instruct20

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