AINotebookcheckabout 1 hour ago
AI Models Fail Overcooked Coordination Test
2 min read
Researchers tested four vision-language AI models on the game Overcooked using a new environment called DuoCook. The strongest model, GPT-6-Astra, completed an average of 26.2 percent of orders, while others scored between 8.8 and 11.2 percent. Adding shared memory improved Claude's score but increased API costs for all models.
Level
Hype check
Tap to vote and see what everyone thinks.
