OpenAI's GPT-5.6 Sol, Terra, and Luna, explained for a small business that just wants to know which one to use

The first message I got about GPT-5.6 was not excitement. It was confusion. A client who runs a small accounting practice wrote to ask which of the three new ChatGPTs she was supposed to buy, and whether the cheap one would "make mistakes on the numbers." She had seen the three names, Sol, Terra, Luna, and read them as three products she now had to choose between and probably pay for separately. That confusion is the most important thing about this launch, because it is going to be common.

So let me clear it up before anything else. You are not being asked to buy three things. OpenAI changed how it names and ships its models, and the change is actually in your favour once you understand it. There is one new generation, GPT-5.6. It comes in three sizes, and the whole point of the three sizes is that you stop overpaying by running every small job on the biggest, most expensive engine.

That is the genuine news here, and it is more useful to a small business than any benchmark. For two years the question was "is the new model better." The question now is "which size of the new model fits this particular task," and getting that answer right is worth real money for a business that uses AI at any volume. This is a pricing and design change dressed up as a model launch, and the pricing part is the part you can act on.

Here is the whole thing in plain terms, then how to choose.

What OpenAI actually changed

OpenAI split one model into a named family of three and gave the naming a logic it did not have before. With GPT-5.6, the number identifies the generation, and the names Sol, Terra, and Luna identify durable capability tiers that can each improve on their own schedule. So a future GPT-5.7 Luna would be the next version of the fast, cheap tier, and you would already know roughly what it is for from the name alone (OpenAI, previewing GPT-5.6 Sol).

The names are deliberately evocative. Sol is the Sun, the flagship and most capable. Terra is the Earth, the dependable mid-tier built for the bulk of real business work. Luna is the Moon, the lightest and fastest. OpenAI announced the three on June 26, 2026, as a limited preview, with access initially restricted to a small group of trusted partners and, per reporting, some government-related constraints on who can use the most capable tier (VentureBeat, June 2026).

Why does a naming change matter to a business that just wants help with its email and its spreadsheets? Because the old approach quietly encouraged waste. When there is one model, you use it for everything, including tasks that a model a fraction of the price would handle perfectly. By making the tiers explicit and naming them by job, OpenAI is nudging every user toward the question that saves money: does this task actually need the flagship, or will the cheap tier do it just as well? For a business running AI at any real volume, that question is the difference between a sensible bill and a silly one (DataCamp, GPT-5.6 family).

Sol, Terra, and Luna in plain terms

Each tier maps to a kind of work, and once you see the mapping, choosing is easy. Sol is the flagship, built for the hardest problems: complex multi-step coding, security research, and the kind of dense reasoning where being right matters more than being cheap or fast. It is the most expensive and the slowest of the three, and for most small businesses it is the one you reach for least, because most day-to-day work is not actually that hard (OpenAI Help Center, GPT-5.6 preview).

Terra is the one most small businesses will live in. OpenAI positions it as the strong lower-cost option for high-volume business tasks: customer support, internal tools, and document analysis. This is the workhorse tier, the model that answers the support ticket, reads the contract, drafts the proposal, and powers the automation that runs hundreds of times a day. It trades a little of Sol's peak capability for a much lower price, and for the overwhelming majority of business tasks you would never notice the difference in quality, only in the bill.

Luna is the sprinter. It is the fastest and most cost-efficient of the three, built for routine, lighter work: summarizing a thread, drafting a quick reply, classifying an email, the simple repetitive automation that needs to happen instantly and cheaply at scale. Where Terra is the workhorse, Luna is the conveyor belt. The art for a business is matching the tier to the task: Luna for the routine, Terra for the substantive, and Sol only for the genuinely hard, which is the same disciplined fit-to-task thinking we walked through for autonomous work tools in our Copilot Cowork, Codex, and Claude Cowork comparison.

What it costs, and the number that matters for small teams

The pricing is where the three-tier design pays off, and the gap between the tiers is large enough to change how you build. Priced per million tokens, which is roughly the unit of text in and out, Sol costs 5 dollars for input and 30 dollars for output. Terra costs 2.50 dollars input and 15 dollars output, half the price. Luna costs 1 dollar input and 6 dollars output, a fifth of Sol's price (OpenAI, previewing GPT-5.6 Sol).

Sit with that spread for a moment, because it is the whole argument for caring about tiers. Running a high-volume task on Luna instead of Sol is five times cheaper on output. For a business automating something that happens thousands of times a month, a customer-support reply, a data extraction, an email classification, that is not a rounding error. That is the difference between an automation that pays for itself and one that quietly eats the savings it was supposed to create. The cheapest tier that does the job well is almost always the right tier.

The discipline this rewards is the same one that separates profitable AI use from wasteful AI use generally. Default to the cheapest tier that produces acceptable quality, test whether a more expensive tier actually improves the result for your specific task, and only pay up when it measurably does. Most businesses do the opposite by reflex, reaching for the most powerful model because it feels safest, and overpay for work the cheap tier handled fine. We put real numbers on that tradeoff in our breakdown of the true ROI of AI agents, and the tiered pricing of GPT-5.6 makes the lesson concrete: matching tier to task is one of the simplest ways to control an AI bill.

You probably cannot use it yet, and that is fine

As of the June 26 preview, you most likely cannot use GPT-5.6 even if you want to. During the preview, Sol, Terra, and Luna are available only through the OpenAI API and Codex to a limited group of trusted partners and organizations with an OpenAI account representative. Crucially, GPT-5.6 is not available inside ChatGPT during the preview, so the millions of people who use the consumer app cannot touch it yet (OpenAI Help Center, GPT-5.6 preview).

OpenAI has said it plans to make the three models generally available in the coming weeks, so this is a short wait rather than a locked door. But the limited preview is a useful reminder not to let a launch headline rush your decisions. The model being announced and the model being usable by your business are two different events, often separated by weeks, and there is no advantage to a small business in being first through the door on day one. The advantage is in being ready to use the right tool well when it arrives.

There is also a small naming-driven sideshow worth knowing about, only so you can ignore it. Because Sol, Terra, and Luna are also the names of well-known cryptocurrencies, the announcement set off a brief flurry of confused crypto chatter. It has nothing to do with OpenAI's models or with your business. If you see the names trending alongside coin tickers, that is the reason, and it changes nothing about which tier you should use for your support inbox.

Which tier fits which job

The practical decision for a small business comes down to honestly classifying your tasks, and most fall cleanly into one of the three buckets. Start with the routine, high-volume, low-judgement work: summarizing messages, drafting first-pass replies, sorting and tagging emails, simple data cleanup. That is Luna territory. It is fast, it is cheap, and the work is forgiving enough that the lightest tier handles it without anyone noticing a quality difference. If a task happens constantly and does not require deep reasoning, default to the cheapest tier.

Then the substantive everyday business work: answering real customer questions, analysing a document, drafting a proposal, powering an internal tool your team relies on. That is Terra, the workhorse. It is capable enough that the output is genuinely good, and priced so that running it at volume does not punish you. For most businesses, Terra will quietly do the majority of the actual work, and it is the tier to standardise on for anything that touches a customer or a decision but is not unusually hard.

Reserve Sol for the genuinely difficult: complex coding, intricate multi-step analysis, security-sensitive work, anything where a wrong answer is expensive and the problem is hard enough that the cheaper tiers visibly struggle. The mistake to avoid is using Sol by default because it is the best, the same way people once bought the most powerful computer for a job that needed a tablet. The skill is not picking the strongest model. It is picking the cheapest one that clears the bar for the task in front of you, and only stepping up when a real test shows it is worth it.

What to do right now

Right now, the move is preparation, not panic. You cannot use GPT-5.6 in ChatGPT yet, and there is no prize for adopting it the hour it goes generally available. What you can do, today, is the work that makes you ready to use it well: list the tasks where your business already uses AI or could, and sort each one by how hard it really is. That sorting is the whole skill the new tiers reward, and it is useful no matter which vendor or model you end up on.

If you build automations on the OpenAI API, the tiered pricing is a direct invitation to audit what you are running. Anything currently using the top model for routine work is a candidate to move down a tier and cut its cost by up to five times with no real quality loss. When GPT-5.6 reaches general availability, you will be able to assign Luna, Terra, or Sol per workflow, and the businesses that have already classified their tasks will capture that saving immediately while others are still reading the announcement.

And if all of this still feels like a lot of names and numbers for someone who just wants their business to run more smoothly, that is the honest and common reaction, and it is exactly the gap we exist to close. You do not need to track every model launch. You need someone to match your actual tasks to the right tools and build it so it runs. The tiers are a detail. The outcome, less time on repetitive work and a bill that fits what you actually use, is the point.

The bottom line

GPT-5.6 is less a leap in capability than a smarter way to package and price it, and for a small business that is arguably more useful than another benchmark record. Three tiers, Sol for the hardest work, Terra for the bulk of real business tasks, and Luna for routine high-volume jobs, give you a way to stop overpaying by matching the engine to the task. The number is the generation. The name is the job. Once that clicks, the choice that used to feel like a technical decision becomes a simple business one.

My accounting-practice client did not need to buy three things, and she did not need to fear the cheap tier on her numbers, because the cheap tier was never going to touch the work that required judgement. She needed what most small businesses need: someone to tell her plainly which jobs go to which tier, and to set it up so the right one runs each time without her thinking about it. That is the calm version of living with AI, one tool doing the routine work quietly in the background while she does the work only she can do. The model names will keep changing. That calm is the thing worth building toward, and it is reachable now.

Sources

• OpenAI — Previewing GPT-5.6 Sol: a next-generation model
• OpenAI Help Center — A preview of GPT-5.6 Sol, Terra, and Luna
• VentureBeat — OpenAI unveils GPT-5.6 Sol, Terra and Luna, limited preview
• DataCamp — GPT-5.6 Sol, Terra, and Luna: OpenAI's Next-Gen Model Family
• EdenAI — GPT-5.6 Sol: Benchmarks, Pricing and API Access Guide
• Andrew.ooo — What is GPT-5.6? Sol, Terra, Luna Explained
• Crypto.news — OpenAI sparks crypto frenzy with GPT-5.6 Sol, Terra and Luna names
• Intellectia.AI — OpenAI Unveils GPT-5.6 Model Names: Sol, Terra, and Luna