<oembed><type>rich</type><version>1.0</version><title>alp wrote</title><author_name>alp (npub175…9g6w0)</author_name><author_url>https://yabu.me/npub175nul9cvufswwsnpy99lvyhg7ad9nkccxhkhusznxfkr7e0zxthql9g6w0</author_url><provider_name>njump</provider_name><provider_url>https://yabu.me</provider_url><html>My evolution in AI/LLM supported development over the last 3 years:&#xA;&#xA;Google Gemini&#xA;&#xA;Terrible, ridiculously expensive, the results were pretty mediocre. And a provider I didn’t like, but back then there wasn’t anything else around. I was constantly nervously checking the Google Cloud Console to keep the costs under control.&#xA;Results: a few crappy little tools.&#xA;&#xA;A brief detour to OpenGPT’s LLM, whatever it was called&#xA;&#xA;Better results, but still ridiculously expensive.&#xA;&#xA;Claude Code&#xA;&#xA;Better and cheaper, with a flat rate of about €100 a month. For a long time, it was dumb as a rock, but the latest models were able to handle a large project.&#xA;&#xA;Z.AI&#xA;&#xA;Chinese models that are much better, without the moralizing, and with a cheaper flat rate of about €55 a month. Fantastic results.&#xA;&#xA;Private provider&#xA;&#xA;This feels like the next level: anonymous, pay as you go with Sats, cheaper overall, and the best models for whatever task. The only minor downside is that the models still talk too much.&#xA;&#xA;And for the last 3 days, I’ve even been using one particular provider for free while they test their infrastructure.&#xA;&#xA;Still: Every model is bad at design and CSS. You have to step in a lot, even with the CSS architecture.&#xA;&#xA;Anyone who hasn’t kept up with this progress and has spent the whole time moaning “AI sucks” has missed out and will have a hard time catching up.</html></oembed>