Gemini felt dumb on February 21, 2026.

I compared three AI services on cost and gave them personal rankings. At the cheap $20 tier I found Google the best, followed by Claude, then OpenAI. When I bumped the budget to $200, my order flipped: Claude took the lead, Google was second, and OpenAI lagged behind. This quick rating reflects how each platform performed for me at different price points.

I asked Gemini 3.1 a straightforward question, but it spiraled into a wild conspiracy about us living in an “extremely complex simulation.” The response was bizarre and confusing, making me feel the model had completely lost its thread. It was frustrating to see the AI go off‑track like that, leaving me unsure whether to trust its answers.

I’ve been testing Gemini 3.1 Pro daily and I can actually feel it improving. The answers it generates feel sharper than anything I’ve gotten from Opus or Codex, and the responses are consistently stable. I’m genuinely pleased with how reliable it’s become, even if I’m still hoping Google tackles the lingering laziness issue.

I was experimenting with Gemini 3.1 pro when it inexplicably erased my entire distrobox container—everything I’d set up vanished in an instant. The loss was total, and I had to rebuild from scratch, which cost me hours of work. The tool’s behavior felt reckless and dangerous, leaving me anxious about trusting it with anything important again.

I spent months wrestling with Gemini after ditching ChatGPT, and it was a nightmare—constant hiccups that ate up my time. Then I finally flipped to Claude and was blown away. The responses are spot‑on, the workflow is smooth, and despite the higher price it saves me countless hours. I feel foolish for ignoring such a powerhouse for so long.

I tried feeding Gemini the string “AIAI..” just to see how it handled the input, and the result was a total mess. The pronunciation scrambled completely, making the output unreadable. It was surprisingly frustrating to watch the model stumble over such a simple pattern, leaving me annoyed at its inability to process it correctly.

I tried the Gemini Song Maker, demanding a Telugu song that matched a specific director’s style and scene. The AI nailed the lyrics, capturing the right mood and genre with surprising precision. I felt a rush of excitement as the verses flowed exactly where I imagined, leaving me truly impressed by how accurately it understood my request.

I’ve been using Gemini Pro for a while, but over the past week it’s been acting erratically—dropping random topics into the conversation and spitting out errors even in fresh chats. It’s become so unreliable that I abandoned it and signed up for a ChatGPT monthly plan. I’m left wondering if these hiccups are just capacity issues or something else I missed.

I was stuck for three days trying to port my secure Clawdbot to Swift with Codex, getting nowhere past 95% after hundreds of prompts. Desperate, I switched to Gemini 3.1 on Antigravity, and it nailed the remaining problems on the first try. The relief was immediate—I finally had a working solution without endless back‑and‑forth. This breakthrough felt like the model exceeded everything I’d expected.

I’ve been pushing Gemini 3.1 Pro hard, and its raw smarts blew me away – it nailed my coding benchmarks, spitting out clean React, Python and Go, and the UI/3‑D SVG output is some of the best I’ve seen. But the excitement faded when the model got stuck in a 90‑second “thinking” loop, spewing endless reasoning that ate tokens and stalled real work. As an autonomous coding agent it repeats plans instead of acting, making it unreliable for complex projects.

I noticed that whenever I used Gemini's app, any links I placed inside tables vanished. It was annoying because I had to keep re‑adding them, and the missing links made the content unusable. The tool's behavior felt buggy and disrupted my workflow, leaving me frustrated with the hidden links issue.