Gemini felt dumb on October 14, 2025.

I’ve been using Gemini 2.5 with the same system prompt for weeks, and it used to work flawlessly. Lately it’s missing obvious words, mis‑understanding my one‑line prompts, and the latency in AI Studio spikes after just 60k tokens instead of 300k. I even hit a rate‑limit warning despite low usage. Trying a new account didn’t help, and I’m left wondering if the API could fix the lag or let me tweak the prompt and temperature, even though I’m not a programmer. The whole experience has become frustrating and unreliable.

I’ve been paying $200 a month for the Codex CLI, but after noticing the slowdown people complained about, I switched back to Gemini CLI. Pairing the two feels like a dynamic duo: Codex gets stuck in loops, while Gemini 2.5 acts like a doctor, diagnosing and un‑jamming it. The combo now works extremely well, and I’m using Gemini way more after its recent update—so impressed I can only imagine what Gemini 3.0 will bring.

I tried the Gemini CLI and was constantly hit with mistakes and garbage code, which made the experience frustrating and disappointing. Compared to Sonnet 4.5, which feels almost god‑like, Gemini felt like a step backward. I’m pleading for Google to improve it so I can abandon Anthropic’s limits, give them my money, and rely on a tool that actually works.

I’ve been using Gemini 2.5 for the last month and it’s been a nightmare – every query feels off, the answers are garbled, and the model often stops working altogether. It’s become almost impossible to get anything useful out of it, and I’m left worrying that this is just a pre‑release wobble before the 3.0 launch. The frustration is real, and I’m not sure if others are stuck in the same rut.

I kept turning on the Gemini app, only to watch it shut down by itself each time, both on my PC browser and iPhone app. I tried restarting it a thousand times, but it just kept going offline, which was infuriating. Compared to Google AI Studio, Gemini felt unreliable and useless, leaving me feeling annoyed and stuck.

I tried to keep a long conversation with Gemini, but now I can’t even load my pinned chat because the prompt history is too big. It’s irritating—ChatGPT lets me reference older messages, but Gemini just blocks me, forcing me to start over. The tool’s limitation feels like a needless roadblock, making the whole experience frustrating and less useful.

I was chatting about orbital velocity and de‑orbit burns and asked about “perigee.” The model nailed the spelling everywhere except one line that said “perigsee,” which made me cringe. That slip felt jarring, especially since it’s a basic term the AI should know, and it left me uneasy about trusting its accuracy.