VLAN - Matched 200 Trajectories - Wheelchair: leaderboard

Independent evaluation cohort. S4T7=30real-worldtrajectories; language/actionfamilymeans andharmonicAccSR. A4T14=fixed200testtrajectories, 2/scene, seed42, nineVLMs, identicalbudget; notinterchangeablewith1931fullsplit. FullsplitDroneSRact/GRinconsistency held separately. Metric: Native success score (0–1). Source: https://arxiv.org/html/2611.18346v1

Metric: Native success score (0–1). Source: arxiv.org. Saturation forecast: Around May 2027. 8 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-4o0.382#754
2Qwen 2.5 VL 3B Instruct0.335#2648
3InternVL3.5-8B0.33#1588
4Molmo2-8B0.28#1629
5Qwen 2.5 VL 7B Instruct0.19#1893
6Qwen 3 VL 8B Instruct0.115#1025
7Llama 3.2 11B Instruct0.095#5226
8Idefics3-8B-Llama30.005#5775

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=vlan-matched-200-trajectories-wheelchair · How It Works · Data refreshed daily, snapshot 2026-11-17.