do AI companies keep training and improving their older models, or is most of the work focused on building new models and tools?
i've been using an older Sonnet model, and some things it used to struggle with are now producing surprisingly good results. makes me wonder how much these models improve over time.
commit 856bd8a
Older models work fine for coding, but the company might discontinue them. Consequently, you will be forced to use newer models.