The latest OpenAI signal is not just that GPT-5.6 Sol is stronger. It is that businesses now need a routing map: which model gets the hard task, which model gets the cheap task, and where the human checkpoint lives.
| Sol | Deep reasoning, complex code, security review | Expensive lane; use for narrow work with review |
|---|---|---|
| Terra | Everyday analysis, planning, operations drafts | Balanced lane; use for repeatable team workflows |
| Luna | Fast drafts, sorting, intake, support triage | Low-cost lane; use for volume and first-pass work |
| Codex | Software changes, tests, docs, site edits | Implementation lane; require change review |
| Human | Approval, policy, money, trust, security | Final gate; never automate away accountability |
The first buyer question is access
OpenAI says GPT-5.6 Sol, Terra, and Luna are in limited preview through the API and Codex for approved organizations, and not in ChatGPT during the preview. That matters because buyers should prepare workflows now, but should not promise staff that a normal ChatGPT switch is already available.
Route the model before you route the task
The clean business split is simple: Sol for high-stakes reasoning, difficult coding, cyber-defense review, scientific work, and long-horizon planning; Terra for balanced daily knowledge work; Luna for high-volume drafting, sorting, classification, and inexpensive support tasks.
Pricing changes the workflow design
OpenAI's help article lists Sol at the highest GPT-5.6 price tier, Terra in the middle, and Luna as the lowest-cost lane. That makes prompt caching, reuse, task batching, and model routing business decisions, not technical trivia.
Codex is the implementation surface
For operators, Codex-style access is where the Sol story becomes tangible: a model can inspect code, propose edits, run checks, and hand the change back for review. The correct buyer move is to attach that power to one workflow with source notes and approval gates.
The human layer is not optional
OpenAI's safety material describes layered safeguards for higher-risk areas. A business should mirror that pattern: source notes for facts, role permissions for tools, and a named human approver before outputs affect money, legal commitments, customer trust, cybersecurity, or published claims.