List your company

Procedural assembly reasoning on LEGO-Puzzles

52Height AccuracyInternVL3-78B
18.7227.363644.64Feb 21, 2026

Evaluation results

MethodLinks
InternVL3-78BAccess=Open-source2026.02526336546480582925223747.3
Gemini-2.5-FlashAccess=Proprietary2026.02525837555874534940462950.1
GPT-4oAccess=Proprietary2026.02496641516587515153724957.7
Qwen2.5-VL-72BAccess=Open-source2026.02435838395776575274434352.7
InternVL3-38BAccess=Open-source2026.02406039555781593347123647.2
Claude-3.5-SonnetAccess=Proprietary2026.02396042486178583749546453.6
Qwen2.5-VL-32BAccess=Open-source2026.02356038524579514566434350.6
TPRU-32BAccess=Ours, Backbone=Qwen2.5-VL-32B2026.02346135475576524870494852.3
TPRU-3BAccess=Ours, Backbone=Qwen2.5-VL-3B2026.0230543240376949143082435.2
Qwen2.5-VL-3BAccess=Open-source2026.0229553036326548191642532.6
TPRU-7BAccess=Ours, Backbone=Qwen2.5-VL-7B2026.02235637404567553149323642.8
Qwen2.5-VL-7BAccess=Open-source2026.0220573247386756292282536.5