semisol on Nostr: Also, on older models, the reasoning effort roughly used to correspond with the API, ...

Also, on older models, the reasoning effort roughly used to correspond with the API, it was only scaled down a bit on low/medium.

Now you can’t get true xhigh or max.