GPT Sol Pro Is My "Go-To" Model
For the stuff I am building and maintaining I find OpenAI’s GPT Sol Pro to provide the best value versus other coding agents.
I prefer it to Claude Opus.
Claude Fable is too expensive. When I used it briefly I was initially impressed, but had to go back and fix some of the coding tasks it had “completed”.
I’ve yet to try Kimi K3 (if I do I will load it into Codex). Most engineers that I have seen comment about Kimi K3 feel that it is not quite on par with Fable nor Sol, and given Kimi’s price (it is not cheap), I probably will not try it anytime soon.
My issue with GLM 5.2 a few weeks ago was that it struggled to understand earnings call transcripts out of the box and would require training.
By the way, we brought OpenAI’s GPT Luna model into Kilby this week to pair with Claude Haiku for tasks that are not complex. I’m working to make Kilby as inexpensive as possible. Eventually I will bring in open source models, but they will require a bit of work before on-boarding both in terms of training and generic tools. I am capacity constrained, so the open source effort will have to wait.
One thing that has not changed for me as it relates to AI usage is that I prefer working in CLI versus cloud interface. This is true for Claude, GPT and Kilby. Working locally in CLI allows for faster model responses and better context awareness (local data ingestion is more complete and better context awareness) and is a better ROI.