Gemini felt dumb on January 24, 2026.

I’ve been testing Gemini 3’s CLI and was pleasantly surprised by how solid it’s become. It actually caught bugs that both GPT 5.2 and Opus 4.5 missed, which felt like a real boost in reliability. I’m still puzzled how it keeps getting better even though the core model hasn’t changed, but the experience has been impressively smooth.

I’ve been using Gemini with a custom system prompt to get answers in Matt Levine’s witty style, and it used to work flawlessly. Over the past week the model spends ages debating whether to follow the style instruction, then ignores it and gives a plain answer. The chain‑of‑thought is filled with meta‑rules instead of the topic, making responses slower and less entertaining. It feels like a clear downgrade in both speed and quality.

I’m hitting the rate limit almost instantly, and GPT‑4o (3) is acting badly – it skips my prompts, ignores instructions and keeps throwing error messages. Every time I try to regenerate I waste generations, and the tool feels unusable. I’m left wondering if this is just a temporary glitch, because right now it’s far too limited to be helpful.

I tried to create an image with the model, but it just wouldn’t produce anything—only a broken preview link showed up. The tool kept stalling, and I kept getting the same empty result no matter how I tweaked the prompt. It was irritating to watch the process spin without delivering the visual I needed, leaving me stuck and needing to find another solution.

I spent a frustrating half‑hour wrestling with Gemini because it kept spitting out images even when I never asked for them. Every prompt turned into a picture, and I tried all sorts of tricks—changing the topic, adding filler text—until I finally asked a simple “How are you today?” and it finally stopped. The constant, unsolicited visuals were irritating and broke my workflow.

I kept trying to open the IDE chat, but it just crashes every time. Even after reinstalling the whole thing, nothing changed. The constant failure made me feel stuck and frustrated, wasting time I could've spent coding. It feels like a critical bug that completely blocks me from using the tool.

I tried using Gemini and noticed it started spamming whole articles into my chats. The responses were filled with irrelevant text blocks, making it hard to get concise answers. The tool's behavior was frustrating and felt like a regression, turning what should be a helpful assistant into a noisy dump of content.

I’m seriously annoyed by the recent tweak to the AI mode – the whole “we can delve” phrase keeps popping up and it’s driving me nuts. I tried using the tool as usual, but every time that line appears it feels like the system has regressed. The experience is frustrating and feels like a step backward, making my workflow irritating.

I spent just a minute and a half using Opus, only to get hit with a six‑day timeout—after already enduring a three‑day block for a similar issue. As a Pro subscriber, being forced offline for days feels punitive and useless, especially when I was only testing a quick prompt. The experience was irritating and left me questioning the service’s fairness.

I spent only 90 seconds with Opus before being hit with a six‑day timeout, after already enduring three days offline. As a paying “Pro” user, I felt blindsided and helpless, watching the platform lock me out without explanation. The sudden restriction was infuriating and completely halted my work, making the whole experience feel dangerous and unacceptable.