GPT-5.6 ships as three tiers. Pick the one that fits the job, not the flagship.

Sol, Terra, and Luna are the same GPT-5.6 family at three levels of horsepower and price, so you can match the tier to how hard the task is. They all share the roughly one million token context window and the February 2026 knowledge cutoff. What differs is capability and cost per token. For a small business running automations, that difference is the entire decision. Here is what each tier is built for.

• Luna is the fast, cheap tier. At $1 per million input tokens and $6 per million output, it is built for high-volume, latency-sensitive work: tagging and routing, first-pass drafts, simple classification, quick edits. When each call is easy but you run thousands of them, Luna keeps the bill small.
• Terra is the balanced everyday model. OpenAI positions it at roughly GPT-5.5-level quality for about half the cost, at $2.50 input and $15 output per million tokens. It is the workhorse that handles most of your day: triage, drafting replies, extraction, summarizing, support answers, the bulk of real automation steps.
• Sol is the flagship. At $5 input and $30 output per million tokens, it is OpenAI's top tier for complex reasoning, long-horizon agentic work, and hard coding. This is the model for tasks where being wrong is expensive and correctness matters more than the price of the call.

OpenAI positions the GPT-5.6 family as its most agentic yet, meaning the models are tuned to plan, call tools, and work through multi-step tasks rather than just answering a single prompt. That capability is real, but it does not remove the need for a human to scope the work and review the result. The question is never whether Sol is impressive. It is whether the task in front of you actually needs Sol, or whether Terra or Luna would do the same job for a fraction of the cost.

Default to Terra, save with Luna, and escalate to Sol only when a task genuinely needs it. That one rule handles almost every situation. Here is how it plays out across the four kinds of work a small team actually automates.

1. High-volume, simple steps, use Luna

Tagging incoming leads, routing tickets to the right queue, generating first-pass drafts, classifying messages, cleaning up form data. Each call is easy, but you run a lot of them, so cost per call is what matters. Luna at $1 and $6 per million tokens keeps that bill small. Paying Terra or Sol prices for work Luna handles fine is money left on the table.

2. Your everyday automation workhorse, use Terra

Triage that needs a little judgment, drafting replies that have to sound right, extracting structured data from messy input, summarizing calls, answering support questions from your docs. This is the bulk of real automation, and Terra is built for it: roughly GPT-5.5-level quality at about half the cost. Make it the model your workflows call by default.

3. Genuinely hard or high-stakes jobs, use Sol

Complex reasoning across a dense contract set, a long-horizon agent that plans and works through a multi-step build, hard coding where a subtle mistake is expensive to catch later. When correctness matters more than the price of the call, Sol's extra rigor pays for itself. This is the exception, not the default, and that is the point.

4. Not sure? Start one tier down and test

When you cannot tell whether a task needs Sol or Terra, run it on the cheaper tier first with a handful of real examples and check the output. Most of the time the cheaper tier passes, and you just saved a multiple on cost. Only move up a tier when you can point to a specific failure the cheaper model produced. Let the results, not the fear of using a weaker model, pick the tier.

The five steps at a glance: (1) route by difficulty with Terra as your default, (2) push high-volume simple steps to Luna, (3) turn on prompt caching and the Batch API to cut the bill, (4) keep a human on high-stakes Sol outputs, and (5) test the tier on your own data instead of guessing.