Listed performance of ~5 points less than o3 on benchmarks is pretty impressive.

Wonder if they feel the bar will be raised soon (GPT-5) and feel more comfortable releasing something this strong.