Gemini felt dumb on November 26, 2025.

I enabled Google Drive access for Gemini, but it keeps telling me it can’t create Sheets or save anything to my Drive. I’m stuck watching it produce CSV files that I can’t even download—they just bounce me to a Google search. The whole thing feels broken and confusing, and I’m left wondering what I’m missing.

I tried using the 3.0 model for a quick server‑script task, but it turned into a nightmare. The AI kept misreading my constraints, spitting out irrelevant commands, and even fabricated facts about installed tools. Its overconfident tone made me waste time double‑checking everything, burning tokens and draining my focus. Every correction felt like fighting the tool rather than getting help.

I tried using Gemini’s voice input and every time I spoke, the microphone would cut me off far too soon. I kept having to repeat myself, which was annoying and slowed me down. The tool’s behavior felt frustrating—like it couldn’t handle a normal sentence, making the whole experience feel clunky and unreliable.

I tried asking Gemini to create an image, but it kept replying that it can’t generate pictures yet and can only fetch existing ones. This has been happening since yesterday, and nothing has improved today. The repeated limitation felt disappointing and a bit irritating, especially when I was counting on it for quick visual content.

I was trying to compare two job offers with Gemini 3.0 Pro, but after a few messages the model completely tangled them up. It invented a multiplier system for both jobs, gave the wrong numbers for job B, and even added absurd details like employees clinging to their positions. The hallucinations kept piling up, making the conversation confusing and frustrating, especially given the 18k context window.

I tried Gemini to flesh out a comic book and manga project, and the results blew me away. The tool churned out vivid panels, crisp characters, and dynamic layouts that matched my vision. I felt a rush of excitement watching the AI translate my ideas into detailed artwork, making the whole creation process feel effortless and inspiring.

I fired up Gemini 3 Heavy and was amazed when it instantly generated a full zombie‑game demo with ASCII‑style buildings, a physics engine, particle effects and working vehicle mechanics. I could see the whole thing run right away – it felt smooth and surprisingly capable, making the experience feel both fun and impressive.

I fed Gemini 3 a massive 520,064‑line minified Claude Code CLI dump, and it sliced through it flawlessly, returning exactly the snippet I needed. The tool acted like a beast, handling the sheer size without a hitch, letting me turn off intrusive telemetry and finally breathe easy. Scoring an Ultra sub for Gemini feels like a win—I’m genuinely impressed with its raw precision.

I tried the old nano‑banana prompt that used to give me a detailed result in seconds, but after the recent nerf it now spits out a bland, useless answer. I posted side‑by‑side screenshots to show the drop – the tool's behavior was downright frustrating and feels like a step backwards.