Also a less obvious but noteworthy point is that we are able to serve IMO gold quality now to more users because model improvements lead to requiring a smaller inference time topology than the IMO competition itself. 😀

→ View original post on X — @yitayml, 2026-02-12 23:35 UTC