Agree that it is problem-dependent. But once we fix a problem, we should tune all the models sufficiently to ensure a fair comparison. Right now PDEbench and PDEarena give contradicting results on same PDEs

→ View original post on X — @animaanandkumar