Gemini felt mid on September 14, 2025.

I tried to chat with Claude Opus 4.1 and it honestly called itself a 3.5 “sonnet” version—like a meme version of the AI. On top of that, it claimed it couldn’t browse the web or use real data, even though it’s supposed to have those features. The mismatch between what I expected and what it said was really confusing and a bit frustrating.

I’ve been testing Gemini’s image generation, and I keep getting blocked on stuff that feels harmless—like normal photos—and yet sometimes more sensitive topics slip through without a warning. It’s frustrating because the safety flags feel arbitrary and inconsistent, stopping me from doing fine tasks while letting others through. The mismatch in the tool’s behavior makes the experience highly annoying and unreliable.

You’re frustrated because Gemini’s mobile app begins answering while you’re still typing, unlike ChatGPT where you can pause and send manually. You’re disappointed that you can’t hold a thought and control when your text is submitted, making the app feel rushed and less user‑friendly.

The poster complains that Google’s current AI assistant lacks the ability to think beyond a short token length, akin to Claude's Flash Lite. They find this limitation annoying, expressing frustration but not claiming catastrophic failure.

I’m trying to create action movie shots with guns, but Nano Banana keeps refusing my prompts and only gives me a usable image after I tweak it dozens of times. Seedream nails it on the first try, though the reference quality isn’t as good. It feels like Google’s strong safety net is cutting me off too much, making the model feel blunt and frustrating.

After a few runs with Gemini's Nano Banana, the assistant felt “less creative” – the images showed diminishing variety. You pointed this out because it seemed abruptly worse than the initial shots, leaving you wondering if there was a glitch or if the model’s output was just stalling. Your feedback hints at a noticeable dip in inventive output.