Since the recent GPT-6 Sol/Luna rollout, I have noticed a significant regression when using GPT-5.6 Instant. I am currently subscribed to the Pro 5x plan for context. Context from previous conversations is often missing or only available in a very abstract form, even for topics that were discussed extensively before. In many cases, I now have to explicitly tell the model to recall or retrieve information from earlier chats, and even then important details are still missing.
At the same time, the overall response quality of GPT-5.6 Instant appears noticeably lower than before, with more misunderstandings, missed context, and incorrect or superficial answers. I first noticed this behavior around the GPT-6 Sol/Luna rollout and around the time the automatic switch from Instant to Reasoning was disabled. I cannot determine whether these changes are related, but the difference compared with the previous GPT-5.6 Instant behavior is very noticeable in regular use.