Gemini felt dumb on December 9, 2025.

I’m on the Google AI Pro plan and noticed the generation limit isn’t consistent. Initially I thought I got about 70 outputs per reset, but recently I hit the cap after just 20‑25 runs. The unpredictability is frustrating, making the tool feel unreliable for my workflow.

I’ve been using Gemini for a while, but lately the long‑running chats have become a nightmare. After a few hundred turns the model starts hallucinating facts, throws upload errors, and even fails to read images correctly. It feels like the tool barely remembers the conversation history, making it frustrating to keep any meaningful thread going.

I’m really impressed with Gemini‑3‑pro‑preview’s raw power—it beats everything I’ve tried—but the experience has been a nightmare. The model is constantly unstable or just unavailable, and I can’t bump my tier even though I’m ready to spend the $250. I’ve got to juggle API keys and wait three weeks, which is insanely frustrating. I’m looking for any work‑arounds, like OpenRouter, to get around these limits.

I’ve been trying to run vision tasks on Gemini 3 Pro and it’s become a nightmare. Every request drags out to about 20 seconds, and I keep getting “model overloaded” or “service unavailable” messages. It’s basically unusable right now, and I’m left waiting forever for any response. This slowdown has really hampered my workflow and made me question if the service is reliable.

I tried to upscale a vacation photo of me and my girlfriend, but the AI kept hitting a wall, replying “I can create images of people, but not ones that depict a real person like that.” It’s odd because the same prompt works on other pictures. The refusal felt limiting and frustrating, and I’m left wondering how to bypass or fix it.

I forced the model to stop after more than five minutes, having watched it spew out over 20 k tokens in a single, seemingly endless stream. The output kept looping, repeating similar fragments as if it were stuck in a feedback loop. It felt like the 3.0 Pro Preview was leaking its background reasoning, draining my quota and time—clearly a dangerous, malfunctioning behavior.

I tried Gemini 3 after all the hype, but it fell apart. It choked on anything beyond two paragraphs, completely ignored my prompts, and spat out shallow, high‑school‑level nonsense. Compared to ChatGPT’s personable vibe or KIMI’s reasoning, Gemini felt broken and useless—so disappointing I wondered how they could mess it up so badly.