Gemini felt terrible on December 4, 2025.

I spent months building a passion project, and every time I used AI Studio or Vertex Studio, the video model got the answer right. But when I switched to the API, it missed every single time, even though I copied the same prompts, temperature, and sent the video as Part.from_bytes. The inconsistency made the whole video‑understanding feature useless, and I’m left frustrated and searching for any fix.

I tried using Gemini for several chats, hoping it would help with quick answers, but every response was nonsensical gibberish. The tool's behavior was maddening—I couldn't get a coherent reply no matter the prompt, and it ended up wasting my time and shaking my confidence in the platform.

I started a fresh Gemini chat and sent dozens of prompts, but after opening a second chat and returning, only the last few messages remained and the model seemed to forget everything else. It was confusing and irritating to lose the earlier context, especially when other long chats kept their full history. This glitch made the experience frustrating.

I tried using AI Studio expecting fresh results, but Nano Banana Pro just spitted out the exact same unedited image over and over, and Gemini 3 Pro flat‑out replied “failed to generate response.” It made the whole workflow impossible—nothing useful came out, and the whole experience felt broken and frustrating, leaving me unable to get any work done.

I spent a lot of time testing Gemini 3.0 Pro and was blown away by its razor‑sharp reasoning, flawless image generation, and killer coding chops—building front‑ends and breaking down concepts felt effortless. But the tool’s creative writing was a nightmare; long‑form pieces were bland and it kept missing key instructions, leaving me frustrated despite its otherwise stellar performance.