I can’t get over the fact how even the smallest GPT 5.6 variant improves dramatically in ability by simply giving it more test time compute. With this release, setting the reasoning slider appropriately for the task is almost more important than picking the correct model variant. In my very limited testing the last couple of hours, I only switched to a bigger model with lower reasoning to get faster results than with a smaller model on higher reasoning. I am sure my model expectations will drastically change over the coming days and weeks, and then I have to use Sol, but right now, Luna on high reasoning seems already quite good.