By skyforbes This could just be some weird placebo effect, but I’m going to ask anyway in case anyone else is experiencing the same thing.
I’m from Italy, and after hitting the 30-day age verification wall, I decided to bypass it by creating a fresh US Google account using a fake GPS app and a VPN on my phone. Now, whenever I use Google AI Studio, I’m tunneling strictly through the US.
Here is the weird thing: The model performance is night and day depending on the region.
When I’m on the US connection, Gemini 3 feels "raw" and uncensored. It’s sharp, creative, and feels exactly like the model on release. But if I switch back to European IPs, the model feels worse, almost like it’s been lobotomized.
It is overwhelmingly noticeable in coding.
On the US account, it writes full, complex scripts and nails the logic on the first try. And when it doesn't, if i ask it to fix it, it finds the issue quickly. On the EU version, it’s just… lazy. It gives me way more issues in the code, and when i ask it to fix them, it never finds the correct fix for them even if i give it the exact error message.
Has anyone else A/B tested this? Is Google routing EU traffic to a "safer" (over-aligned) checkpoint to comply with the AI Act while the US gets the full weights? Or am I just imagining things?
