Gemini felt dumb on December 10, 2025.

I tried using Gemini 3 to read screenshots of my math notes for finals, but the model keeps missing the images or spitting out bizarre hallucinations (see the video). It’s frustrating because Gemini 2.5 handled the same task flawlessly weeks ago. I’m stuck—either I find a workaround or I revert to GPT, which I’d rather avoid.

I tried Gemini Live after the latest update and was instantly struck by how natural it sounded. It kept my accent intact, delivered longer, more coherent answers, and didn’t end every reply with that nagging “anything else?” question. The conversation flowed smoothly, making the experience feel less like a bot and more like talking to a helpful companion.

I started using Gemini 3 for scientific writing and was impressed at first, but lately its answers feel rushed and shallow. The tool throws out cookie‑cutter sentences, overloads me with em‑dashes even when I ask to avoid them, and often lacks any real rationale. It’s much faster than GPT‑5.1, but the quality has dropped, leaving me frustrated and uneasy about hallucinations. The only bright spot was a single NBPro diagram that was surprisingly detailed.

I’ve been testing Gemini for work and it used to handle detailed, even graphic content without flinching, just giving neutral responses. Since the 3.0 update it’s turned ultra‑sensitive—now it blocks anything, even a cleaned‑up version, and even flags the same text that was fine yesterday. The sudden censorship is frustrating and stalls my workflow.

I signed up for the Ultra Plan expecting high‑resolution PNGs around 5 MB, but every image I get is a tiny 300 KB JPEG and, oddly, lacks the watermark. It feels like the service is underdelivering on the promised quality, leaving me frustrated and questioning whether I'm even on the right tier. I'm asking if anyone else is seeing the same downgrade.

I’ve been using Gemini for creative writing and usually start fresh chats to keep the context window manageable. After two days in the same conversation, the model suddenly lost all memory of the previous discussion. It’s frustrating because a week ago I could maintain a week‑long thread without issues, but now the context just vanishes, making the tool feel unreliable.

I use Gemini every day, but each new chat wipes the slate clean, forcing me to re‑explain my projects and preferences over and over. It’s been a huge hassle, especially after switching from other AIs where I could export memory. I tried pasting a context doc manually, which helped, then built a tool (mindlock.io) to automate the process. The constant need to re‑seed the model feels frustrating and inefficient.

I tried uploading a batch of PDFs to Gemini and it handled up to five without any hiccups, but the moment I added a sixth file everything broke. The tool just stopped responding or gave garbled output, which was really irritating because I needed to process a larger set at once. It felt like hitting a wall mid‑task, turning a simple job into a frustrating scramble.

I spent hours trying to get the AI to work, but every attempt hit the same “Error: went wrong. Please try again later” message. I verified my age, cleared cookies, switched browsers, used incognito, changed Wi‑Fi and even moved locations, yet nothing changed. Waiting weeks didn’t help—just endless frustration and wasted time.

I tried the new Gemini template for a Veo commercial about r/Bard. I ran three videos with different presets: Shiny, Space, and pixelated. The first one was okay but missed the promotion bit, the second felt the best with both the promo and space vibe, and the third basically failed. Still, the later videos were more consistent, so it was a mixed experience.

I dug into Seedream 4.5’s API for weeks, stitching together JSON calls to see where it really shines. The tool consistently spat out crisp typography, lifelike product shots, and steady character looks across dozens of references – all in 4K. I was impressed by how smooth the workflow felt for scripted pipelines, even if it struggles with wild stylization. The experience left me excited to plug it into my own projects.

I was drafting my undergraduate thesis and vented to Gemini that the writing was bad and the topic felt silly. To my surprise, Gemini read my dedication, recalled it, and crafted an emotional appeal back to me—even though I’d never discussed personal stuff before. I was impressed by its nuance, yet I’m uneasy about how deeply it’s parsing my words. The experience felt both clever and a little unsettling.