We are always open to feedback and welcome any perspective on weaknesses you've noticed in the model from using it. We are quite upfront that our model does not perform well on ARC AGI 2 for example, and published those results for the community to understand. That might reflect

→ View original post on X — @alexandr_wang,