I mostly use Gemini for everyday Q/A and research type stuff. I find it's pretty accurate and gets straight to the point. I mostly use Claude and very recently

I mostly use Gemini for everyday Q/A and research type stuff. I find it's pretty accurate and gets straight to the point. I mostly use Claude and very recently Codex for systems software dev. I'm very interested to see what changes. I'm wondering how these models are getting better at understanding and generating code. Are they being trained on more data because these companies use their free tier customers' data?