Skip to content

Comment on GPT-6 Astra in code review: Gains, privacy, and costparent

Comments

Same experience here, but I have some strange feeling.

Sol I'm used to working a month ago doesn't feel the same I'm using today, slower and less accurate. My gut feeling is that they quantize previous models to prioritize new ones and, who nows, make the new one look better.

Up to July I was using mostly anthropic models and the feeling was the same, so much so that I was able to predict every model release 1 or 2 days before public announcements.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.