I have an open source grant from OpenAI so I use gpt 6 Astra light for all my coding stuff. I have a 3-4 pinned sessions which are 1 per project backed by a GitHub repo and deployed to a VPS where they will either use APIs for Claude, gpt6 or DeepSeek via atlascloud.

I previously had an open source grant for Claude and that was quite good as well. There is a sort of "just good enough" point that we're reaching where all the providers are on similar levels. Like cloud. Everyone reached feature parity with AWS. Now it doesn't matter.

I don't use Gemini. Google is great. I use AI mode in search for general knowledge stuff but honestly the thousand moving parts and unique ways in which their products break is annoying. I used to work at Google. They have amazing tech. But I came in through acquisition. I'm not a die hard engineer. I think they built from the ground up and everything is limited by the scope of the infrastructure. So they design APIs like that. Google itself was a single input box. Simple. But I think Gemini and the API needs work. They are going to do well as cloud infrastructure for enterprise. And their models will probably get widely deployed in all their own products.