OpenAI started rolling out GPT-5.6 to the public on July 9, 2026 — and for the first time, the company isn’t shipping one model. It’s shipping three. Sol, Terra, and Luna are the same GPT-5.6 family split into three tiers of power and price, and the gap between the cheapest and the most expensive is wide enough to swing your monthly AI bill by 5x for the exact same work.
For a small business, that changes the question. It’s no longer “should we use the newest ChatGPT?” It’s “which of the GPT-5.6 tiers should we point at each job so we get the quality we need without overpaying?” This guide breaks down what launched, what each tier costs, and a simple rule of thumb for routing your work — plus three settings that can quietly cut your bill by more than half.
This is written for owners and operators, not engineers. You don’t need to understand the model internals to make good calls here — you just need to know which lever to pull for which job. Let’s start with what OpenAI actually shipped.
What actually launched: the three GPT-5.6 tiers
GPT-5.6 arrived as three named tiers. They share the same core model family but are tuned — and priced — for different jobs:
- Sol is the flagship, built for the hardest reasoning: complex multi-step agents, scientific and technical work, and serious coding.
- Terra is the everyday workhorse. OpenAI positions it as roughly matching last generation’s GPT-5.5 quality at a lower price — the tier most businesses will live in.
- Luna is the lightweight, high-volume option, made for simple, repetitive tasks: sorting, tagging, routing, and first-pass drafts at scale.
The rollout was notable in its own right: GPT-5.6 shipped only after a U.S. government national-security review, part of a new rule requiring the most powerful models to be submitted for review before public release. OpenAI has been moving fast all year — from its GPT-Live voice assistant to this three-tier launch. For business users, the practical takeaway is simple: all three GPT-5.6 tiers are available now through ChatGPT and the API.
What each GPT-5.6 tier costs
| Tier | Best for | Input / 1M tokens | Output / 1M tokens |
|---|---|---|---|
| Sol | Hardest reasoning, agents, coding | $5.00 | $30.00 |
| Terra | Everyday business tasks | $2.50 | $15.00 |
| Luna | High-volume, simple tasks | $1.00 | $6.00 |
A “token” is roughly three-quarters of a word, so one million tokens is about 750,000 words in and out. The headline is the spread: Sol costs five times what Luna does. On the same volume of work, that’s the difference between roughly a $25 day and a $125 day. Most businesses don’t need their most expensive model for every task — and that gap is exactly where the savings live.
Which GPT-5.6 tier should your business use?
The mental model is simple: match the model to the difficulty of the job, not to the newest name. Here’s how the three GPT-5.6 tiers map to real work.
Use Luna for high-volume, simple work
If a task is repetitive and the right answer would be obvious to a sharp intern, Luna is your tier. Think tagging support tickets by topic, sorting inbound leads, pulling fields out of forms, generating first-draft product descriptions, or classifying reviews as positive or negative. You’re paying $1/$6 instead of $5/$30, and for this kind of work you won’t notice a quality difference.
Use Terra for everyday business tasks
Terra is where most small businesses should set their default. Writing and editing marketing copy, drafting emails and proposals, summarizing meetings, answering customer questions with nuance, and light analysis all sit comfortably here. It’s close to last year’s top model in quality but priced for daily use. When you’re not sure which tier a task needs, start with Terra and only move up if the results disappoint.
Save Sol for genuinely hard problems
Sol earns its price on complexity: multi-step research, financial modeling, debugging real code, or running an autonomous agent that has to plan and self-correct across many steps. If a mistake is expensive or the reasoning runs deep, pay for Sol. For everything else it’s overkill — and overkill is what runs up an AI bill.
How to tell which GPT-5.6 tier a task needs (a 10-second test)
When you’re staring at a task and not sure where to send it, ask three quick questions. First: is the answer basically mechanical — the kind of thing a careful assistant could do without much judgment? That’s Luna. Second: does it need real writing quality, tone, or a bit of reasoning, but nothing exotic? That’s Terra. Third: would a wrong answer cost you real money, or does the task involve deep, multi-step reasoning, code, or an agent acting on its own? That’s Sol. Ninety percent of the time you’ll land on Luna or Terra — and that’s the point.
A quick example: what routing saves a real business
Say you run a 10-person e-commerce shop and use AI for three things: writing product descriptions (high volume, simple), answering customer emails (everyday), and deciding which products to reorder (harder). Send all of it to Sol and you pay top rates across the board. Route the descriptions to Luna, the emails to Terra, and only the reorder analysis to Sol, and you’re paying the premium rate on a thin slice of the work instead of all of it. Layer caching on your product catalog and batch the descriptions overnight, and the same workload can cost a fraction of the all-Sol bill — with no drop in quality where it counts.
Three settings that quietly cut your AI bill
Picking the right tier is half the savings. These three levers are the other half — and most businesses leave them on the table.
Turn on prompt caching (up to 90% off repeat context)
If you send the same background information over and over — brand guidelines, a product catalog, a standing set of instructions — caching lets the model reuse it instead of re-reading (and re-charging for) it every time. Cached input is discounted by roughly 90%, with a short minimum cache life of about 30 minutes. For any workflow built around a big, stable block of instructions, this is the single biggest lever you have.
Batch non-urgent work (up to 50% off)
Not everything needs an answer in two seconds. If you’re generating 500 product descriptions overnight or working through a backlog, the Batch option trades speed for a discount of up to 50% on both input and output. Run it while you sleep and wake up to half-price results.
Route by task, not by habit
The most expensive habit is sending every request to your fanciest model. Decide the rule once: simple and high-volume goes to Luna, everyday goes to Terra, hard and high-stakes goes to Sol. Many tools built on the API let you set the model per workflow. In the ChatGPT app, just use the model picker deliberately instead of defaulting to the top option every time.
ChatGPT app vs. the API: which are you actually paying for?
This distinction matters. The prices above are API rates — pay-per-token — which is what you pay if you or your tools call OpenAI directly. If your team simply uses ChatGPT through a monthly subscription, you’re not billed per token, but the same logic still applies: use the lighter, faster tier for routine questions and reserve the heavy model for hard ones, because even on a subscription the top tier is slower and rate-limited. The businesses that feel the per-token math most are the ones building AI into their own apps, automations, and customer-facing tools — which is also why it’s worth auditing the software you already pay for to see where those costs are hiding.
What GPT-5.6 means for the tools you already use
Here’s the part that’s easy to miss: you may benefit from GPT-5.6 without changing anything. A growing share of the business software you already pay for — help desks, CRMs, writing assistants, marketing platforms — runs on OpenAI’s models under the hood. As those vendors upgrade to the new tiers, your existing tools quietly get faster and cheaper to run, and some of that saving tends to reach customers over time. It’s worth asking your key vendors which model they’re on and whether the upgrade changes your plan. If you’re actively shopping, our roundup of the best AI writing tools for small business is a good place to compare.
Your 2026 GPT-5.6 routing cheat sheet
- Sorting, tagging, routing, bulk first drafts → Luna
- Everyday writing, email, summaries, customer replies → Terra
- Deep research, coding, financial modeling, agents → Sol
- Repeated background or system prompts → turn on caching
- Overnight or bulk jobs → use batch
Set these defaults once and you’ll capture most of the savings without thinking about it again.
Frequently asked questions
Is Terra good enough to replace GPT-5.5?
For most business tasks, yes. OpenAI positions Terra as comparable to GPT-5.5 in quality at a lower price, which is why it’s the sensible default for everyday work.
Do I have to pick one GPT-5.6 tier for everything?
No — and you shouldn’t. The whole point of three tiers is to route different jobs to different models. Most businesses will lean on Luna and Terra constantly and call on Sol only when a task truly needs it.
Will using a cheaper tier hurt quality?
Only if you point it at work that’s too hard for it. Matched correctly — simple work to Luna, everyday work to Terra — the output is effectively the same, and you keep Sol in reserve for the tasks that genuinely need it.
Three moves to make this week
- Audit where you use AI now and label each task simple, everyday, or hard.
- Set Luna as the default for everything on the simple list, and Terra for everyday work.
- Turn on caching for your single biggest repeated prompt, and move any bulk job to an overnight batch.
None of this requires new software or a developer. It’s mostly a matter of being deliberate about which of the GPT-5.6 tiers does which job — and once you’ve set those defaults, the savings keep compounding month after month. If you only do one thing, start by caching your largest repeated prompt: it’s the fastest way to see a smaller bill without touching quality.
The bottom line
The three GPT-5.6 tiers are good news for small businesses, because they finally let you stop paying for capability you don’t use. The winning move isn’t picking the smartest model — it’s matching each job to the cheapest tier that does it well, then switching on caching and batching for the repetitive work. Do that and you can get better results than last year while spending less. To go deeper on controlling AI costs, read our breakdown of the 2026 tokenmaxxing spending crackdown and our guide to auditing your SaaS stack — and if you’re still deciding where AI belongs in your business at all, start with when to hire versus use AI.