This year became a strong point of decline for Gemini. Since the release of version 2.0, they had started to dominate and could compete with Claude and GPT without problems. All competitors became very worried.
But in 2026 too much changed, and I will simply describe a series of mistakes that were made, where auto-moderation is not the only problem:
- Studio limits became terrible; the global compute shortage affected this too;
- the Gemini website received such awful system instructions that it works worse than even free Chinese models, even if you use their flagship model;
- the external censorship filter went through several rounds of strengthening, and now it simply deletes the entire message after generation;
- model 3.1 seemed not bad, but it gives shorter answers and more interpretations, often agreeing with the user too much; they apparently wanted to reduce token usage, but this made the answers awful;
- a low-quality, negligent decision with the recommendation to connect a paid subscription to Studio — for some people it still does not work: people bought the subscription, but did not receive the limits;
- the release of model 3.5 Flash is a move downward. It is even dumber than 3.1 Pro, writes even more briefly, and messes up even more.
Versions 2.0 through 3.0 were a magnificent solution for analysis, discussions of complex topics, and creative tasks — from writing articles to books. The current flagship 3.1 and 3.5 Flash are absolutely unsuitable for this.
I created my own benchmark for testing models on creative tasks, and starting with version 3.1 these tasks began to sag, while competitors were getting better and better.
Previously, what stopped me from using GPT was that it agreed too often, and Claude because it had a context-window limit. Now these problems are gone. GPT is literally ready to argue with you and refute any of your claims on the paid plan, while Claudia magnificently fits large texts for analysis and gives a consistent, reasonable result without hallucinations.
Since around March, I have almost stopped using Google AI Studio services, only occasionally looking in and comparing it with other solutions.
I became interested in how much such problems scare users away and where people went for creativity. This topic, and many neighboring ones, confirm what I discovered: Google kept constantly ruining its Studio, often does not know how to implement its own tools, and, while having the best arsenal and market offer, always made some strange, insane moves.
It is worth remembering that they still never managed to implement their Nano Banana for a wide range of users — on the Gemini website it always catches double watermarks, Studio closed access, and external sites attached payment to the API, gaining profit where Google itself could have dominated.
In programming, Gemini could compete only in rare moments, so in agentic tasks they failed to enter the market of Claude and GPT. And their agentic system is made as well as AI Studio — that is, damn badly, lagging and so on.
And so these guys decided that the only place where they held confident, solid positions — creativity and analysis — needed to be “fixed.” They “fixed” it. It no longer works.
Thanks to everyone for trying to convey this information.
But now Google Studio has a new icon in browser bookmarks. 
Let’s congratulate them on this wonderful release and this important update!