I was using it today and the responses just feel a bit off compared to how it was performing right before the update.
I don’t have quantitative benchmark data or side-by-side metric logs to prove it (this is strictly based on my personal experience using it today) but the outputs just feel noticeable less solid than they were right before the update. The overall depth, nuance, and logic flow seem a bit off for my usual prompts.
Has anyone else noticed this, or is it just my specific use cases today? Would love to know if there are any known workarounds or if this is on the radar.
Hi,
Yes I noticed a quality drop as well, I test just about every model that comes out and I think 3.6 is a disaster. 3.5 flash lite however does seem to be an improvement. I ran head to head tests between 3.6 and 3.5 flash lite in one shot coding and flash lite produced results that are of a much higher quality. Also 3.6 seems to generate a whole lot of rubbish that has little or no relevance to the job.
Hi, do you have 3.5 flash lite in Antigravity? Because i do not have.
I added a new cms feature to a customer site building with 3.6 flash medium. the speed was excellent and reviewing step by step it seems to do less loops, but for some reason i had to continue to press for testing even when then plan had it. It was back to “faking it” which was common 6-7 months ago. I ended up adding new rules and a skill to fix the behavior. overall, it was very obvious that my quota was much less impacted for the same work as the past.
That said, until AGY gets the ability to pop sub-agents using different models, I am still using flash 3.6 as a single agent for each activity or request. AI Studio has no problem running lesser models as subagents or course.
Yes, but only for the first two hours, just after they deployed the models, while using conductor. After various Antigravity restarts and time passed, it started to behave like before with the benefits of 3.6. The team probably messed something up during deployment and fixed it later.
What quality drop. I cant even run antigravity. It freezes when starting. Google AI is done. I will switch to anthropic!!!
