GPT-5.2 and GPT-5.2-Codex got a 40% speed boost with a new inference. That was fast

→ View original post on X — @testingcatalog