VLAN - (a) Humanoid - Instruction Agreement: leaderboard
Full1931trajectorysplit; user-conditionedlanguageinstruction. SRlan requiresalluser-specificsemanticconstraints; IAS weightedstructuredsimilaritytocanonicaltarget. Seven releasedVLMs; exactInstructlabelsretained; noactionSRderivedmetricused. Metric: Native score (0–1). Source: https://arxiv.org/html/2611.18346v1
Metric: Native score (0–1). Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.2 11B Instruct | 0.846 | #5226 |
| 2 | Molmo2-8B | 0.84 | #1629 |
| 3 | Qwen 2.5 VL 7B Instruct | 0.781 | #1893 |
| 4 | Qwen 2.5 VL 3B Instruct | 0.752 | #2648 |
| 5 | Idefics3-8B-Llama3 | 0.75 | #5775 |
| 6 | Qwen 3 VL 8B Instruct | 0.742 | #1025 |
| 7 | InternVL3.5-8B | 0.738 | #1588 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=vlan-a-humanoid-instruction-agreement · How It Works · Data refreshed daily, snapshot 2026-11-17.