Never use GCP / #Google models for long term model use, unless you're an enterprise customer with contracts
basically:
>announce 3.5 flash
>3.5 flash output token price is 6x more expensive than 3 flash preview
>not token efficient, smart but logic still sonnet level, overeager to use tools
>still not a significant leap over gemini 3
>faced backlash for benchmarks vs real world quality gap
>deprecate 2.5 flash model and 3 flash, true price performant workhorse model
You can no longer access it if you create new gcp project
You can still access it but if you didn't create a new project but its volatile
if you want 'flash' model, you don't, your only option is 3.1 flash lite which IMHO not as good even 2.5 flash
If balanced cost:performance workhorse and long term access really matters, use open weights model, or models that is not gemini... typical google doing google things behavior
rough to rely on gcp ecosystem, google's track record is to kill their products
#gcp #gemini #google