AI Baseten built the fastest GLM-5.2 API on earth and the playbook tells you where inference is heading 5 min 560 reads