Gemini felt dumb on January 10, 2026.

I tried running a text‑based game in Gemini, then reloaded the same JSON prompt into ChatGPT to see how the two handled the status output. Gemini spitted out a massive dump with every engine detail, which felt a bit overkill. ChatGPT trimmed it down, giving just the essentials I needed to keep playing. The contrast was clear – Gemini was verbose, ChatGPT was concise – and that helped me decide which format fits my workflow.

I tried using Gemini 3 Pro in Google AI Studio to analyze a batch of 106 color images, hoping the perfect prompts I generated would fix everything. Instead, it kept misdescribing the colors, and even single images often got wrong results unless arranged in a grid. The tool’s behavior was consistently off‑track and frustrating, making the whole task feel impossible.

I added the custom system prompt to Gemini’s settings and immediately felt a lift in interaction quality. The model became more honest and accurate, acting like a critical‑thinking partner that actually uses up‑to‑date info. I’m now trusting Gemini more because it stays focused, gives concrete next steps, and feels motivating without being overly clinical.

I was just brainstorming a light‑hearted cozy story when Gemini unexpectedly dumped a bizarre “A Sensitive Query” response that felt like hardcore porn. I was aiming for something low‑stakes and fun, but the AI completely mis‑interpreted my prompt, delivering an off‑topic, inappropriate draft. The whole experience was confusing and frustrating, leaving me doubting the model’s ability to handle simple creative requests.

I tested the 1M‑context Gemini on several PRO accounts and was stunned by how useless it was. It couldn’t even open a simple 20‑page PDF or any other file type, and it crashed around 85k tokens. The experience felt like a slap in the face—getting a model that promises so much but delivers practically nothing, leaving me frustrated and angry.