Gemini felt dumb on February 5, 2026.

I’ve switched almost entirely to 5.3‑codex for my daily workflow, and it completely eclipses Gemini 3. The old, barely‑used Gemini setup feels obsolete now. I wish the Gemini CLI team would stop polishing the UI and actually try to compete—right now they’re just watching Anthropic and OpenAI race ahead, which is incredibly frustrating.

I signed up for Gemini hoping it would boost my research, but it’s been a nightmare. Every time I try to discuss an image it claims it can’t see it or spouts hallucinated details, and the canvas feature constantly drops information. Even in AI Studio, a simple question triggers random code changes. The tool’s behavior feels broken and wasteful, making the subscription feel pointless.

I’ve never seen this AI stumble before, but today it let me down three separate times. Each failure interrupted my workflow and left me scrambling to fix the mess it created. The repeated breakdown felt maddening and risky, turning what should've been smooth assistance into a source of frustration and wasted time.

I spent hours testing free AI face generators, and every “perfect” portrait felt synthetic—too smooth, plastic‑like. I’d churn out dozens of images just to snag one that barely seemed natural, turning the creative process into tedious maintenance. Switching to a setup that prioritises texture and consistency gave me a more believable result, but I’m still curious what subtle clues might still betray its AI origin.

I was checking out the new Anthropic Opus 4.6 rollout and couldn't help but notice its massive context win—256K and 1M tokens with recall rates of 93% and 76%. In stark contrast, Gemini 3 is stuck at 24.5% and 45.4%, which left me feeling quietly defeated. The constant push to “quantize everything to save money” just adds to the frustration, making the tool feel like it’s lagging far behind the competition.

I tried writing a sprawling, intricate story using Gemini, but as the narrative grew, the model began dropping characters, plot points, and details one by one. It felt like the AI was slowly losing track of the world I’d built, forcing me to constantly remind it of earlier events. The experience was frustrating and made the writing process feel disjointed.

I’m really excited about KLING 3.0 and have been testing it nonstop on Higgsfield. The new version feels smoother, and the AI video generation model finally hits the sweet spot between speed and quality. I dove into the best‑use cases, and the results kept impressing me—each clip looked sharper and the workflow was far less glitchy. Overall, the update made my experiments feel productive and enjoyable.