Please add GPT-5.6 Sol Pro to Codex as an asynchronous “Consult Pro” or “Pro Review” mode.

What variant of Codex are you using?

cli

What feature would you like to see?

I am not asking for every Codex tool-use turn to run on Pro. That would be unnecessarily expensive and slow. Instead, Codex should be able to package the relevant context and send one high-value question to Pro for an independent, long-running evaluation. The intended workflow is: 1. Codex performs the normal investigation and implementation work. 2. When it encounters a high-value uncertainty, architectural decision, difficult root-cause problem, mathematical question, or final adversarial review, the user can explicitly invoke “Consult Pro.” 3. Codex prepares a controlled context package containing: - the final goal and success criteria; - the current system or code state; - confirmed facts and primary evidence; - current hypotheses and their weak points; - failed attempts and complete errors; - constraints that must not be violated; - relevant files, diffs, logs, and test results; - one clearly stated decision or question. 4. Pro independently reconstructs the problem instead of accepting the current agent’s conclusions. 5. Pro distinguishes facts, inferences, and unknowns; searches for counterexamples and alternative explanations; and returns a concrete decision plus the smallest useful verification experiments. 6. Codex receives that response and validates it through local tests rather than blindly following it. The Pro consultation should run asynchronously and may take 20–120 minutes. Codex should be allowed to continue unrelated work while it runs. Preserve a Pro conversation ID so the user can make two or three valuable follow-up requests without rebuilding the entire context. This can have a small, separate quota—such as a limited number of consultations per month, per-use credits, or a maximum runtime per request. Show the expected quota cost and require explicit confirmation before starting. Max and Ultra do not replace this capability. Max is deeper reasoning along the current path, while Ultra is parallel delegation. Neither necessarily provides the sustained, independent, globally coherent evaluation that Sol Pro can provide for problems that cannot be cleanly decomposed. What Codex needs is not “Pro for every agent turn,” but an official, asynchronous, traceable, separately metered Pro consultant channel.

Additional information

_No response_