Finally. Something new from OpenAI. Just switched out the models for Subtxt and the inference speed difference is insane! Won’t know about the reasoning until after a few days, but already happy to see some improvements.