OpenAI reports 400B-token tests of Fable 5.2

The author shares a month-long hands-on evaluation of Fable 5.2, highlighting its exceptional intelligence and capability to generate complex browser-based applications from single prompts, such as a voxel engine and a voxel-based Earth simulator. However, the author notes significant usability drawbacks, including extreme slowness, frequent overly cautious behavior resembling a strict parent, and poor performance in multi-agent setups compared to single-agent configurations. Open article →