Sol is six operating points, not one number
OpenAI exposes none, low, medium, high, xhigh, and max reasoning effort. Independent evaluations show a real quality curve across those settings, but the latency curve is just as real. This is why the profile consolidates every Sol row: they are configurations of one product decision, not six separate model pages.
A useful production policy starts at the lowest effort that can pass the task. Escalate on verifier failure, uncertainty, or task class. That policy captures much of Sol’s upside without paying max-effort latency on work that never needed it.