Game2Real-Bench: game-to-real visual reasoning correlation hits 0.935 across 25 VLMs
Qingan Liu · hf · 2026-10-19
Game2Real-Bench (Dalian University of Technology & NTU) tests whether visual reasoning measured in toy games transfers to real-world tasks, across fine-grained perception, spatial reasoning, constraint reasoning, and sequential planning. With 1,000 verifiable game samples and 400 aligned real-world samples, evaluations of 25 VLMs show strong game-to-real consistency (Pearson/Spearman 0.935). One-shot transfer improves constraint reasoning but hurts fine-grained perception. A 24K-sample dataset is also released.
More from Research
- Embedding arithmetic demo: cartoon − animation − comic = diagram — mayfer · 2026-10-23
- Omnii Genome: Single Generalist Model Outperforms AlphaGenome Across All Benchmarks — JosephJacks_ · 2026-10-23
- Domingos Cites New Result Claiming RL Is 'Basically Useless' vs Supervised Learning — pmddomingos · 2026-10-23
- Word Arithmetic on Embeddings: "Bathroom Minus Toilet Equals Kitchen" Beats SAE Probing — mayfer · 2026-10-23
- Wolfram Opens Rule 30 Prizes to AIs: $30,000 for Three Open Problems — cephaloform · 2026-10-23
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23