Gemini 3.0 Flash Preview thinking effort deterioration since +-18 Feb

Not sure if there was a change made to the routing algorithm or thinking token algorithm.

Since late last week and including today. Auto mode is giving worse answers, if I run my scripts again then the output is generally better.

It used to be good always on the first try and has only been acting up the last few days.

Suspect it’s trying to preserve tokens or compute and giving lazy answers.

1 Like

So an update it’s still sadly doing this today. I’m going to try set thinking budget manually and hope this fixes it.

Consistently its giving bad answers on the first attempt and perfect answers on the second attempt it’s extremely weird.