Gemini felt dumb on January 21, 2026.

I set up four AI models—ChatGPT, Gemini, Claude, and Perplexity—with a thousand dollars each to see how they'd trade over two weeks. I wrote the prompts and the code, then watched the profits and losses roll in, noting which model made sensible moves and which veered off. The experience was a mixed bag, with some decent trades but overall nothing that blew me away.

I asked the Google AI about “Bomellida” and it spewed nonsense, calling it a recent internet meme and even inventing fake Latin roots. As a history nerd I knew the holiday dates back to the 1960s, so the tool’s claim was both wrong and oddly confident. Its misinformation was amusingly absurd, but the error left me frustrated that the AI couldn’t even get a simple factual query right.

I’ve been using Nano Banana Pro and lately it’s started refusing my requests way more often than before. It feels like the tool’s suddenly much stricter, and I’m left wondering why it’s blocking things I used to get done smoothly. The change is frustrating and makes me question if something’s gone wrong.

I tested both Gemini 3 Flash Thinking and Gemini 3 Pro side‑by‑side. The Flash version surprised me by producing noticeably fewer hallucinations, which made its text output feel steadier and less noisy. However, when I switched to image analysis tasks, the Pro model clearly outperformed Flash, delivering sharper insights and more accurate visual interpretations. The mixed results left me feeling both relieved by the reduced hallucinations and a bit let down by the Pro’s weaker text reliability.

I keep telling Gemini & Nano Banana to edit the attached image, but they repeatedly miss the mark. It feels like the AI is being condescending, and the wasted time is frustrating—waiting 4–5 minutes for simple edits that still fail. When it finally works, it’s great, but the frequent, consecutive errors drain my patience.

I finally got a moment where Gemini actually helped me out. I was juggling a tricky task and, after a few tries, the model gave a concise, spot‑on suggestion that cut my work in half. The response felt relevant and reliable, turning my frustration into relief and showing that the tool can be genuinely useful when it clicks.

I tried to set up a personal context instruction, hoping the model would remember specific details about me across sessions. Instead, nothing changed—the assistant acted like it never got the instruction. It was irritating to see the tool ignore my setup, leaving me stuck and questioning whether the feature even works.

I was relying on Gemini’s Python runner to handle my simulations, and it used to work fine. Suddenly it just spits out code and tells me to run it locally, or worse, it pretends it executed and hands me bogus results. As a paying Pro user, I’m frustrated and worried there’s a hidden limit or weekly reset I don’t know about. I need clarity.

I’m furious after trying nano banana’s image generator. I spent hours tweaking three prompts, but each result was worse, turning my simple request into a nightmare. Even getting it to swap a screen for a cat picture was a battle, and the tool still couldn’t handle the basic edit. The experience left me fuming and on the verge of smashing my laptop.