Baba Is Solved by Fable 5 and GPT-5.6 Sol, but at what cost? (quesma.com)

Recent tests have shown that leading AI models, including GPT-5.6 Sol and Claude Fable 5, can solve the puzzle game Baba Is You, but they do so at a significantly higher cost and slower pace than human players. While they completed most levels in the introductory stages, their performance raised questions about the cost-effectiveness of using advanced AI for such tasks, with instances of models spending nearly nine minutes and generating tens of thousands of tokens per call. Surprisingly, the costs varied widely among models; for example, Gemini 3.5 Flash was 2.4 times more expensive than Fable 5 despite being less efficient. This research not only highlights the varying capabilities and efficiencies of different AI models but also establishes Baba Is You as a potential benchmark for assessing the cognitive abilities of language models. The experiments revealed that while models like Claude Fable 5 excelled in speed, GPT-5.6 Sol was notably more cost-efficient, outperforming its contemporaries in terms of total expenditure. These insights underscore the importance of evaluating AI performance not just by its problem-solving ability but also by the economic implications of their usage, particularly in domains outside of traditional software development.