Gemini felt dumb on September 9, 2025.

After an initial week of buggy behavior that quickly improved, Gemini Deep Think delivered impressive, in‑depth plans and itineraries, often generating 50‑100k+ token responses. When the quota was raised from 5 to 10/day, the output depth dropped, and by early September the tool started thinking only 5‑10 minutes per request, producing content that was a tenth of the length. The user feels this nerf has ruined the model’s ability to provide exhaustive analysis and notes that earlier workarounds no longer work.

Nano Banana delivered noticeably good results, with characters that looked close to the originals. I found the output improved when I stripped backgrounds, allowing the AI to focus on the character. Minor details needed tweaking, but overall it was helpful and performed well.

I’ve noticed Gemini’s outputs getting more erratic lately. The system keeps hallucinating facts and giving vague, nonsensical answers—more than usual. Each time I ask it to explain something, the response drifts off-topic, mixing up details and offering no reliable insights. It’s frustrating because I expected a consistently accurate assistant.

I’ve been using Gemini 2.5 PRO daily and noticed that its performance fluctuates each day. Today the model kept generating deep, recursive “thinking” steps, making it hard to get straight answers. The accuracy felt lower than in recent weeks, forcing me to iterate many times to arrive at a correct response. I’m wondering if the model version or setting changes from day to day.

I joined a nano banana hackathon and discovered that public Gemini API apps in AI Studio are charging my account without permission. The TypeScript builder doesn't support a “bring-your-own-key” setup for users, which poses a serious risk of unexpected, heavy charges if the app gains popularity.

I’m frustrated because Gemini keeps throwing wild analogies at me that actually make the concepts harder to grasp instead of clarifying them. Even with good responses overall, the constant “garage” or “bike” comparisons feel overdone and confusing. I’ve had to keep telling it to stop, and I’m looking for a way to disable or reduce those analogies so I can get straight answers.

I’ve been using Gemini’s Code Assistant Premium for two months and suddenly it’s barely working. It stops after just a couple of responses and says my quota’s gone, even though I’m not doing anything big. The context left indicator dropped to only 100k tokens, so the replies are much shorter and ...

I tried changing my hairstyle but Gemini keeps blocking my edits every time, no matter which version I use. Even other AI tools refuse to let me make these changes, so it feels like a huge obstacle. The system is just not allowing it, which is frustrating.

Salom keling sevishamiz lekin xotinim bor

I used to see those “AI is dumb” screenshots and thought they were ridiculous, but lately I've noticed a real improvement. The responses are clearer, more accurate, and I actually get what I ask for—plenty more helpful than before.

Menga sun'iy intellekt asosida rasimlarni sifatini aniqlash madelni tuzib ber

I got frustrated with Gemini for losing my context after a laptop sleep. I’d started editing my novel, paused, turned the computer off, and when I came back the chat didn’t remember where I left off. Paying $20 felt pointless when the tool just forgot what I’d asked.

I’m blown away by Gemini 3.0 CLI—it’s seriously slick, even better than Codex or Claude for code. Google’s delayed launch but finally it hits the mark. I can’t elaborate, but it feels like a game‑changer for developers.

I tried teaching me machine learning basics using both Gemini and ChatGPT. Gemini explained things step-by-step and made sure I understood the fundamentals, while ChatGPT jumped ahead and assumed I knew more than I actually did. Overall Gemini was much more beginner‑friendly.