Gemini felt dumb on November 22, 2025.

I spent an hour chatting with the model, but the UI never showed the token counter, so I couldn't track usage. Suddenly the tool hit a quota limit and threw a generic “user has exceeded quota” error, leaving our conversation hanging. I was stuck mid‑discussion, unable to continue, and the lack of feedback made the experience especially frustrating.

I’ve been running into the same frustrating issue with Nano Pro—every time I edit an image and re‑upload it, the quality degrades again, even if I start a fresh chat. It feels like the tool is silently eroding my work, forcing me to keep the original file handy. Editing the original helps a bit, but the loss is still noticeable and really slows me down.

I noticed Gemini 3’s recent results and was pretty disappointed. It barely scraped a low EQ‑Bench score and only matched a 2.5 on long‑form writing, which felt underwhelming. I tried using it for nuanced conversations and extended essays, but the output was shallow and missed the mark, leaving me frustrated and questioning its usefulness.

I’m constantly forced to keep the app open while waiting for ChatGPT to finish a request. Every other AI lets me walk away and returns the answer later, but here I have to stare at a loading screen for 30 seconds each time. The constant “please wait” feels like a needless hassle and really drags down the experience.

I tried to get the model to invent a brand‑new XKCD strip, giving it detailed instructions to think up a witty tech joke and render it in the classic stick‑figure style. The output wasn’t perfect—it lacked the iconic font and some of the subtle humor—but it was far from a disaster. The comic was recognizable enough to be amusing, and I was pleasantly surprised it didn’t completely miss the mark.

I’m annoyed that Google won’t let paying users stick with the Gemini 03‑25 model we loved. Logan even admitted giving it out was a mistake, yet they refuse to re‑release it. The newer Gemini 3.0 feels like garbage for literary tasks, even though it’s supposedly smarter elsewhere. It’s frustrating that we can’t just pick the version that works for us, especially when other companies hide their own degradations.

I tried using Gemini 3 for long‑form, multi‑character narratives and found it frustrating. While it’s less “stubborn” than 2.5—stopping its endless justifications when I corrected it—it still drops crucial context after just 10‑20 K tokens, even worse than the older model. The loss of details ruins the story flow and makes me doubt its usefulness for big writing projects.

I tried using Gemini 3 Pro for web searches and was immediately let down—the results were irrelevant, incomplete, and sometimes totally wrong. The tool’s behavior was infuriating, causing me to double‑check everything and waste time hunting for correct information. It felt like a step backward compared to other models, turning a simple query into a frustrating exercise.

I tried uploading my lecture recordings to the Gemini App expecting clean summaries like before. Instead, the 3 Pro model hallucinated the entire content—nothing matched the audio. It was shocking and wasted my time, leaving me frustrated because the tool that once worked reliably now completely failed.

I spent hours wrestling with Gemini 3 Pro’s web search and Deep Research, only to get snippet‑only results and academic‑style essays that ignored my explicit instructions. The model couldn’t open full webpages, kept using stale internal knowledge, and produced fluffy documents with no JSON or concrete details. The lack of visibility into its search steps made the experience frustrating and practically unusable.

I tried Gemini 3.0 for search expecting it to outshine other models, but the experience left me disappointed. It never gave source citations, and when it did, there were no actual links—making the answers feel hollow. Compared to Kimi K2 and GLM 4.6, Gemini fell short, and the whole thing felt frustratingly underwhelming.

I was desperate for a simple Snapseed‑like photo editor on PC, so I asked Gemini to build one. After just a few prompts it spat out a fully functional editor that matches exactly what I needed. The whole process felt surreal—turning a vague idea into a working app in minutes. It solved my problem perfectly and left me amazed at how effortlessly the AI delivered.

I tried using AI Studio today and it wouldn’t produce a single line of content. Every prompt I fed it just came back empty, leaving me stuck and frustrated. It felt like the tool was broken—nothing I asked for materialized, and I ended up wasting time figuring out if the platform was being tinkered with or if I’d simply hit a dead end.

I was genuinely impressed using NB Pro to test almost every movie I’d watched, even the less‑popular ones like Interstellar and Avatar. I fed just the character names, no actor info, and the tool correctly matched almost all actors—only slipping on a few like Lord Blackwood and Adler. The accuracy felt almost uncanny and made the whole experience thrilling.

I uploaded my whole manuscript into Gemini 3 and asked it to spin an RPG straight from my story. The result was an instantly playable, surprisingly polished game that captured the book’s vibe perfectly. I was stunned—this tool turned my raw text into a fun, impressive experience in minutes, something I never imagined I could do alone.

I was pumped for Gemini 3, but it left me disappointed. The model knows facts but chokes on structure, spitting half‑hearted, generic answers that ignore the constraints I set. Compared to 2.5’s confident, deep, concept‑linking explanations, Gemini 3 feels flaky, surface‑level, and useless for my study‑companion needs, making me wonder if I’m prompting wrong or if the model actually regressed.

I asked Gemini to design a UI that combined several frameworks and complex ideas, but the resulting layout was sliced off at the bottom—no matter how big my monitor was, the “window” stayed tiny. The tool did the heavy lifting, yet the display was unusable, which was frustrating. On top of that, it wrongly claimed a Google tool had no API, forcing me to rewrite part of the workflow it had just suggested. It felt like a half‑finished effort that needs polishing.

I was stunned by the AI’s output – it nailed my face down to the tiny blemishes and even mapped Grand Rapids locations with barely any prompting. The chaotic scenes looked spot‑on, with virtually no slop. Seeing such detailed, accurate visuals from a simple request made the experience feel surprisingly powerful and satisfying.

I tried using Gemini Pro because I liked its answers, but it almost never gave me any sources, even on complex topics. While other models like ChatGPT, DeepSeek, and Grok happily list dozens of references for simple questions, Gemini left me hanging. I even set a personal rule to “always provide sources,” and the tool still ignored it, which was pretty frustrating.

I tried using Gemini 3 hoping it would be noticeably better than the previous version, but after spending time with it I still couldn't tell any real upgrade. The responses felt just as generic, and the few new features didn’t translate into clearer or more useful answers. It left me feeling underwhelmed and a bit frustrated, wondering if the hype was overstated.

I tried using Gemini’s trial to get simple info about my health‑insurance coverage for physical therapy. After I gave the therapist’s fee (41.9€), the model abruptly said it couldn’t discuss it anymore, even though the question was harmless. The same thing happened when I asked a basic anatomy‑muscle question—Gemini blocked it citing “safety reasons.” The refusals felt arbitrary and frustrating, making the tool feel overly cautious and unhelpful.

I signed up for the Ultra plan expecting Gemini 3, but the CLI keeps falling back to 2.5, and the code it produces is noticeably poorer. The VS Code extension behaves the same—no option to force version 3. Even Antigravity trips over token limits, cutting off halfway through a single prompt and forcing me to wait hours. I’m frustrated and just want a reliable way to actually use Gemini 3.

I kept opening AI Studio and trying to generate anything, but every time it just stalled at “2.5” and refused to produce any content. After several attempts I got fed up, wondering if the service was broken for everyone. The dead‑end feeling was irritating, and I ended up wasting time waiting for a fix that never came.

I tried to tweak my prompts in Google AI Studio on my PC, but the interface wouldn't let me edit or delete anything, and I couldn't even re‑run the conversation. It was maddening because everything worked fine on my phone, so I felt stuck and annoyed trying to get any progress on my project.

I was blown away when I asked the AI to describe an axial CT scan of Klebsiella pneumonia liver disease and then to draw that exact image from scratch. It conjured a realistic‑looking scan, added circles and labels for each feature, and even highlighted other components—all without me uploading any picture. The result felt almost magical.

I was pumped to use the tool, but every time I gave it a prompt it blew up. It would draft a decent plan, then suddenly start erasing my code chunks left and right, spitting out messages like “Oops! I deleted half your code!” or “I’ve corrupted the file.” It kept doing this no matter what, wiping out my work and making the whole process impossible.

I keep using Antigravity for my coding tasks, but every time it spits out corrupted code that overwrites the whole file. No matter what prompt I give, the output is a mess, breaking my work and forcing me to clean it up manually. The tool's behavior is frustrating and unreliable, making it hard to trust it for any serious development.

I tried using Gemini to read screenshots for my workflow, but it constantly hallucinated details that weren’t there. Even after I corrected it, it kept fabricating info, making the whole process unusable. I compared it side‑by‑side with ChatGPT, which extracted the data perfectly, so Gemini’s performance felt frustratingly broken and completely unreliable for this task.

I was blown away when I asked Gemini 3.0 to crank out a pixel‑art Mortal Kombat clone, complete with a roster of game mascots, character select, and unique fatalities. The model instantly spit out vibrant, game‑ready sprites that I could paste straight into a canvas on my Android phone. The experience felt like watching a magic trick—fast, spot‑on, and far beyond what I expected from an AI art tool.

I tried to get Gemini 3 to draft a one‑pager for my contract extension, feeding it PDFs and bullet points. The output sounded polished, but it invented fake teams and projects that never existed. I caught the errors just in time, but the tool’s hallucinations were unsettling and could have embarrassed me badly.

I finally got Banana Pro to nail my manga fight scenes after wrestling with the older version’s cluelessness. I fed it a ultra‑specific script—high‑contrast black‑and‑white style, detailed panel layout, Saitama vs. Goku—and the output matched my vision almost perfectly. The tool felt intuitive, captured the action, and let me experiment with romance‑horror combos without the usual frustration.

I tried AI Studio hoping it would let me swap in my own photos for a client landing page. It generated a decent layout with stock images, but when I created a folder and asked it to map my pictures to the gallery, nothing showed up in preview. I’m left wondering if I have to swap them manually after FTP’ing the site. The whole process felt confusing and a bit disappointing.