VideoGameBench: Can Vision-Language Models complete popular video games?

4 wertyk 1 5/29/2025, 11:36:11 AM arxiv.org ↗

Comments (1)

msgodel · 20h ago
That's pretty interesting. Last year I messed with hooking language models up to inform 6/zmachines (both RL training and inference and even generating games) and have noticed most are awful at navigating and reasoning about graphs. Maybe this changed with the latest ones.