Gemini felt terrible on January 2, 2026.

I tried using Gemini 3.0 for role‑play and brainstorming my novel, but the safety filter kept flagging even harmless lines like “Can the Queen keep up.” Every time the filter fired, the conversation was wiped, tagged as a Sensitive Query, and earlier context vanished. After a few incidents in one week the tool became unusable for creative writing, leaving me stuck with either a broken chat or a switch to another model.

I was used to hitting ‘Repeat’ in Gemini’s Nano Banana and getting a fresh, unique image each time. Lately it’s stuck spitting out almost identical pictures almost 90% of the time. Now I have to clutter my prompts with words like “random” or start a whole new chat just to see any variation—way too inconvenient and frustrating.

I’ve been trying to work with Gemini in AiStudio, but lately it’s been a nightmare. Every time I load a file larger than 100k tokens it just crashes—nothing gets read, and the model forgets earlier context. Coding sessions stall, and I can’t rely on it at all. It feels like the tool has completely broken, making every task frustrating and unproductive.

I’ve been relying on Gemini (Gems) daily, and lately it’s been a nightmare. It can’t pull from its own knowledge base or even read files I upload, and it starts hallucinating wildly, pulling in irrelevant Google Workspace data instead. The drop from its 03‑25 peak is stark—now it literally refuses to access any of my files, making it unusable and incredibly frustrating.

I tried using the Gemini CLI to edit my ~600‑line modules, but it kept spitting out duplicated code across files. The tool would rewrite sections repeatedly, turning a tidy module into a messy mess I had to clean up by hand. This behavior makes the CLI practically unusable for me, and I’m left wondering whether the problem lies with the Antigravity wrapper or the underlying model.

I kept getting unwanted YouTube thumbnails and autoplay at the end of the AI’s replies, which was really irritating. After testing many prompts I discovered that just inserting a YouTube link—even in markdown—triggers an embed. I worked out a workaround by formatting links without punctuation or parentheses, which stops the auto‑embed. I also found custom instructions work better than user‑wide ones for blocking this behavior.

I tried using Gemini to control my smart lights, hoping for seamless automation, but the AI completely messed up the setup. It sent wrong commands, flickered the bulbs, and even caused some to burn out, leaving the house in darkness. The tool's behavior was not just buggy—it was hazardous, turning a simple convenience into a frustrating and potentially dangerous nightmare.

fotgot every thing and response wrong question

I tried using Google’s help‑desk AI on a tech support page, hoping for quick assistance, but it was useless. No matter what I typed—“I need help” or any Google‑related request—it kept replying, “I don’t understand, could you simplify?” It felt like the bot had a single parameter and was deliberately half‑hearted. The experience was maddening and made me doubt even Gemini’s future improvements.

3 pro

I tried using Gemini for my workflow and kept hitting a weird block: after I upload any document and ask about it, the model suddenly refuses to read any image I add later. Uploading the image first works fine, but once other files are in the chat it just won’t analyze the picture. I’ve reproduced it every time, posted screenshots as proof, and it feels like a major, overlooked bug that makes the tool nearly useless for me.