AI models solve Baba Is You levels on new benchmark, but at high cost

GPT-5.6 Sol and Claude Fable 5 solved nearly all levels in the first two stages of Baba Is You on the Baba Is Harbor benchmark, but took significantly longer than humans. Gemini 3.5 Flash was 2.4x more expensive than Fable 5 to solve the introductory stage, and GPT-5.6 Terra was 2.9x more costly than Sol.