Tibo @thsottiaux

We’ve made GPT-5.3-Codex-Spark about 30% faster. It is now serving at over 1200 tokens per second. More to come on speed across the board.