This is for off-the-shelf models. With fine-tuning/post-training, open-source models beat closed source in a lot of specialized tasks/areas

→ View original post on X — @clementdelangue