After the recent update to add new models and remove new ones, a lot of the models have degraded output quality. So far I have mostly used GPT 5.4, GPT 5.5, and Sonnet 4.6 and all three have become worse in interpreting what the prompt was asking and outputting results that are much worse in quality than the same prompt from before the update.

The update removing older (and cheaper) models like 5.1 made it harder to conserve credits but also somehow degraded quality of models.