MulRobBench: leaderboard

Metric: Semantic protocol-decision score (%; aggregate over constraint extraction, target disambiguation, safe action, abstention or reobservation and collaboration trigger; 3,024-sample strict evaluation set of real UAV observations with protocol constraints, degradations and operator language; controlled semantic scores in [0, 1] x100). Source: arxiv.org. Saturation forecast: Around 2032. 17 models tracked.

Top models

#ModelScore
1Qwen 3.5 9B49.78
2Qwen 3.5 4B48.88
3Gemma 4 E4B46.29
4InternVL3.5-8B45.79
5Qwen 2.5 VL 7B44.99
6Gemma 4 E2B43.49
7Qwen 3.5 2B43.27
8aya-vision-8B38.61

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