Gemini felt dumb on December 27, 2025.

I tried asking Gemini about “nano banana” and it confidently replied there was nothing by that name, even though a product called nano banana pro exists. I had to force it to look up the correct term, which felt wasteful. It’s unsettling that most users will just accept the first answer, not knowing about knowledge cutoffs, and I’m uneasy about trusting these models widely until that gap is fixed.

I spent hours prompting Nano Banana Pro to create a plain dark navy background, specifying zero texture, no noise, and crisp vector curves, but every generation turned out with a grainy, CRT‑like mess. Even after 60 tries and dozens of prompt tweaks, the tool kept spewing noisy images, which felt incredibly frustrating and like a clear miss on what should be a simple task.

I tried using Nano Banana Pro to generate a batch of 20 images, hoping it would follow my reference pictures. Instead, it only hit the mark about seven times and completely ignored the rest, which was really disappointing. I feel the tool’s behavior is unreliable and frustrating, and I'm wondering if others are seeing the same problem.

I tried the new Gemini 3 Flash and was instantly impressed. The responses were quick, spot‑on, and felt surprisingly human‑like. I could bounce ideas off it, and it kept up without lag, making my workflow smoother. It’s such a boost that I’m hoping Google won’t throttle it—this tool’s performance genuinely thrilled me.

I spent ages puzzling over a simple conversion, double‑checking my work because the AI seemed to say I’d messed up. After five minutes of frustration I finally saw the answer at the bottom confirming I was right all along. The tool’s misleading hint made the whole process unnecessarily aggravating.

I use Google’s Gemini 3 Pro and Flash daily and find them generally solid – they follow instructions and handle everyday tasks well, even beating GPT‑5.1/2 in my experience. But the small details trip me up: logic gaps, missed requirements, and subtle bugs that only surface later. When I paste Gemini‑generated code into Claude Opus 4.5, Claude instantly spots multiple bugs, explains them, and fixes everything in one go. Claude’s tight instruction‑following and low hallucination rate saves me huge time, turning what takes me ten prompts with Gemini into one or two with Claude. So Gemini feels decent for quick, general work, but it’s still a tier below Claude for serious, precise coding.