On 26 June 2026, OpenAI unveiled its new GPT-5.6 family, split into three models: Sol, Terra and Luna. Behind the names lies a simple, useful idea for an SME: one model no longer fits every job. The smart move is not paying for the most powerful one, but choosing the AI model matched to each task. Here is how to read this lineup without jargon, and how it can cut your bill.
In short
- OpenAI announced GPT-5.6 on 26 June 2026, in three versions: Sol (the most powerful), Terra (mid-range) and Luna (the fastest and cheapest) (source: OpenAI).
- Prices range from about $1 to $30 per million words depending on the version, a fivefold gap between Luna and Sol (source: OpenAI).
- For an SME, the point is not maximum performance but the right task-to-model match: reserve Sol for complex cases, send volume to Terra and Luna.
- At launch, the models are only available to a limited group of partners; general availability is announced for the following weeks (source: OpenAI).
- GPT-5.6 also improves prompt caching (reusing repeated instructions), a concrete lever to lower costs.
What OpenAI announced
On 26 June 2026, OpenAI opened a preview of GPT-5.6, presented not as a single model but as a lineup of three models sharing the same technical generation, at different levels of power and price. According to OpenAI, this family advances software engineering, computer use (an agent that drives applications), knowledge work, scientific research and cybersecurity.
The novelty is less a leap in performance than a catalogue logic. Rather than forcing an expensive model onto every request, OpenAI invites you to pick the right level for the need. It is exactly the decision a leader already makes for other resources: you do not assign a senior expert to sort the mail.
Key takeaway
The real news in GPT-5.6 for an SME is not raw power but choice: three price levels finally let you pay each task at its fair cost, instead of one flat rate for everything.
Sol, Terra, Luna: what each model is for
The three models carry names that deliberately avoid numbers. Here is their positioning, as described by OpenAI.
| Model | Profile | Intended uses | Indicative price (per million words) |
|---|---|---|---|
| Sol | The most capable | Hard problems: complex coding, security research | $5 input / $30 output |
| Terra | Balanced | Business volume: customer support, internal tools, document analysis | $2.50 input / $15 output |
| Luna | The fastest and cheapest | Everyday tasks: summarising, drafting, routine automation | $1 input / $6 output |
Source of prices and uses: OpenAI. The "input price" covers the text you send (your question, your documents), the "output price" the text the model generates in reply. A million words is roughly three to four large novels: for most SMEs, monthly use stays well below that.
The price gap between Luna and Sol reaches a factor of five on output. Using Sol to summarise emails would therefore mean paying five times too much for an equivalent result. That is the whole point of a lineup: matching spend to actual difficulty.
How an SME picks the right model
The right reflex fits in one question: how hard is the task really? A repetitive, well-framed task does not need the most expensive model. Complex legal or technical reasoning justifies it.
List your use cases
Rank by difficulty
Assign a model
Measure then adjust
This tiered approach avoids the most common mistake: wiring your whole flow to the flagship model "to be safe." In practice, the majority of an SME's requests (summaries, drafts, sorting) run perfectly well on the entry-level model.
One model for everything
The right model per task
The quiet lever: caching
GPT-5.6 introduces more predictable prompt caching, with explicit cache breakpoints and a 30-minute minimum cache life (source: OpenAI). In plain terms: if you often resend the same starting instructions (your company context, a how-to, a catalogue), the model can reuse them without charging full rate again.
For an SME automating the same kind of reply hundreds of times a day, this mechanism noticeably cuts the "input" share of the bill. It is a technical detail, but that is often where the real cost of high-volume automation is decided.
Availability is limited at first
Important point: at the 26 June 2026 launch, Sol, Terra and Luna are only accessible to a limited group of partners and organisations, via the OpenAI API and the Codex tool. OpenAI says it is targeting general availability in the following weeks (source: OpenAI). So there is no rush for an SME to migrate: the right move now is to prepare your use-case map, so you can switch calmly when general access opens.
This gradual rollout is a reminder of a prudent rule: a model announcement is not immediate availability. Better to test on a real case before overhauling your tools.
FAQ
Is GPT-5.6 available for my business today?
Not yet for everyone. As of 26 June 2026, access is limited to a small group of partners. OpenAI announces general availability in the following weeks (source: OpenAI). Prepare your use-case map in the meantime.
Should I always pick the most powerful model?
No. That is the costliest mistake. Reserve Sol for genuinely complex tasks (code, legal analysis, numeric reasoning). Most SME uses (summaries, drafts, support) run fine on Luna or Terra, at a fraction of the price.
What does it actually cost for an SME?
It depends on volume, but the price gap runs from $1 to $30 per million words depending on the version (source: OpenAI). Most SMEs use well under a million words a month. By assigning each task to the right model, the bill stays very contained.
Do I need to change tools to benefit?
Not necessarily. Many tools you already use (assistants, automations) rely on these models behind the scenes and will switch to GPT-5.6 when it opens. What matters is understanding the lineup logic to steer your costs.
Conclusion
GPT-5.6 marks less a revolution in power than a change in method: AI is now sold as a lineup, and the right purchase depends on the task. For an SME, value comes not from the most expensive model but from a thoughtful match between real need and model level. Start by mapping your uses, assign the cheapest model that does the job, then move up only where quality demands it.
To go further on choosing a model fit for your business, browse our other LUWAI Mag resources or see how other leaders made the leap in our success stories.
