GPT-5.6: the official guide to choosing the right model and automating without overpaying

OpenAI just published its official guide to building agents with GPT-5.6. Behind the marketing showcase lies a piece of information that matters for small businesses in the Basque Country: the best model is almost never the most expensive one. And that's exactly what SMEs who automate poorly tend to forget.

GPT-5.6 is three models, not one

Announced on July 9, 2026, the GPT-5.6 family replaces the "one model does everything" approach with a three-tier offering. OpenAI even changed its naming system: the alias gpt-5.6 now points to gpt-5.6-sol, the flagship model, but there's also gpt-5.6-terra for everyday work and gpt-5.6-luna for high volumes at low cost.

• Sol: the flagship, built for long reasoning, complex code, and tasks that demand maximum autonomy.
• Terra: the balanced model, designed for everyday team work — writing, analysis, support.
• Luna: the most cost-effective model, built for high volumes and simple tasks repeated thousands of times.

On July 30, 2026, OpenAI cut Luna's price by another 80% and Terra's by 20%. The message is clear: the industry wants businesses to stop paying for power they never use.

The numbers that change the game

The benchmarks OpenAI published are worth reading closely, because they say something very concrete for a business that wants to automate:

• On Agents' Last Exam (55 professional domains), Sol scores 53.6, 13.1 points ahead of Claude Fable 5. But even at medium reasoning effort, it beats its competitor by 11.4 points for roughly a quarter of the cost.
• Terra and Luna outperform Fable 5 at roughly one-sixteenth of the cost. One-sixteenth. Not 10% cheaper: 16 times cheaper.
• On the code index, Sol (maximum reasoning) sets a new record at 80, using less than half the output tokens, less than half the time, and roughly a third of the cost.
• On Terminal-Bench 2.1 and DeepSWE, Sol beats previous records on command-line workflows and long-duration engineering tasks.

Translation for a business owner in Anglet, Bayonne or Biarritz: you probably don't need Sol. You need the model that matches your task, and that will often be Terra or Luna.

The real question for a small business: choosing without overpaying

OpenAI's official developer guide stresses a point most sales pitches leave out: the model choice must come after defining the scope, never before. In practice, for a small business in the Basque Country looking to automate, that means three questions to ask:

1. What exactly is the task?

An agent that sorts emails, fills in a spreadsheet, or answers common customer questions doesn't need deep reasoning. It needs speed and controlled cost: that's a job for Luna. An agent that analyzes complex quotes, compares contracts, or prepares tailored sales proposals: Terra. An agent that codes an entire application, manages a technical project, or reasons for hours on a business problem: Sol — and even then, only for genuinely complex cases.

2. How often does the task repeat?

The economics here are blunt: if you run an agent 1,000 times a month, every cent of price difference per call becomes a real budget line. That's where Luna becomes your best friend. Conversely, an agent run 20 times a month can afford a more powerful model: the total cost stays negligible, and quality goes up.

3. What happens with mistakes?

OpenAI's guide recommends putting guardrails in place, using structured outputs (strict JSON formats) and systematically checking results. For a small business, that boils down to one simple rule: an agent working unsupervised on important decisions is a design flaw, whatever model you chose. AI should handle the repetitive tasks, and human validation should stay on the decisions that carry real weight.

The technical updates that matter to professionals

OpenAI's builder guide details three mechanisms useful for serious automations:

• Programmatic Tool Calling: the agent can write and run small programs that coordinate its tools, filter intermediate data, and only surface what matters. On tool-heavy tasks, this sharply cuts down the number of round trips and the token cost.
• Multi-agent: one instance of GPT-5.6 can coordinate several sub-agents in parallel and synthesize their results. That's the principle behind "ultra" mode, which launches four agents by default for demanding tasks. In plain terms: instead of one agent doing everything sequentially, several agents split the work.
• Structured outputs: always output the same strict JSON format, so you can plug the AI into your business tools without surprises. That's the difference between a fun demo and a production tool.

None of this requires being a developer. Automation platforms are gradually building these mechanisms in, and that's precisely where a small business can get ahead: those who automate early, with the right method, pull further away from their local competitors.

In concrete terms, in the Basque Country

Let's look at three typical examples of businesses in the region:

• A restaurant in Bayonne: bookings, frequently asked questions, review management, last-minute reminders. A Luna agent, connected to the Google Business page and the inbox, absorbs 80% of repetitive requests. Monthly cost: a few euros. Time saved by the front-of-house team: several hours a week.
• A tradesperson in Biarritz: quotes, follow-ups, job tracking. A Terra agent, fed with past quotes, prepares consistent proposals that the tradesperson approves in two minutes. It doesn't replace their expertise, it removes the paperwork.
• A communications agency in Anglet: monitoring, first drafts of content, client reporting. A multi-task agent that compiles figures, drafts first versions and prepares materials. The team keeps creative direction, the agent absorbs the volume.

In all three cases, the same logic applies: identify the task, pick a proportionate model, keep human validation on what matters. That's the heart of OpenAI's guide, and it's also the method we apply with businesses across the region.

Traps to avoid

• Paying for Sol when Luna is enough: the "most expensive = best" reflex sends API bills through the roof without improving the result.
• Automating without guardrails: an agent that writes emails in your name or handles your accounting needs clear limits and human validation.
• Ignoring data quality: the best model in the world can't make up for illegible quotes or a poorly maintained customer database.
• Switching tools every month: workflow stability matters more than the latest release. Migrate when there's a measurable gain, not out of fashion.

What to remember

GPT-5.6 isn't just another update: it confirms that AI is becoming a multi-tier service, where performance is paid for on demand. For a small business in the Basque Country, that's excellent news: automating no longer costs a fortune. You still need to pick the right tool, scope the task properly, and keep humans at the center.

At Mister Anderson, we support small and mid-sized businesses in Anglet, Bayonne and Biarritz through this transition: diagnosing which tasks can be automated, choosing the right tools, setting up guardrails, and training teams. People first, AI for the rest.

Sources: OpenAI — "GPT-5.6: Frontier intelligence that scales with your ambition" (July 9, 2026), OpenAI Developers — "Building agents," "Model guidance" (accessed August 14, 2026).

Sol, Terra, Luna: the decision table

This table isn't an absolute truth — every business has its own use cases. But it's a solid starting point. Most small businesses starting out with automation instinctively aim too high, on the reflex that "more expensive means safer." The opposite is true: the real risk isn't under-provisioning, it's overpaying for a model whose capabilities will 90% go unused.

How to get started without getting burned

For a small business in the Basque Country, automation doesn't start with buying a subscription or writing code. It starts with an honest inventory of your day-to-day:

1. List your repetitive tasks: the ones you do every week without thinking about it — entering quotes, replying to standard emails, updating spreadsheets, following up with clients.
2. Quantify the time spent: two hours a week here, three there. An automation that saves an hour a week is worth it if it costs less than a few dozen euros a month.
3. Start small: pick ONE task, the simplest and most frequent one. An agent that sorts incoming emails and drafts standard replies is an excellent first project.
4. Define the validation rules: decide what the agent can do on its own and what requires a human look. Write it down. That's your guardrail.
5. Measure: after a month, compare the time saved against the real cost. If the gain is there, roll it out on the next task.

This gradual method avoids the two classic mistakes: buying an over-powered tool that never gets used, and rolling out mass automation that goes wrong for lack of supervision.

Frequently asked questions

Do you need to be a large company before automating?

No. It's actually the opposite: small businesses have simpler processes to automate, volumes that are easier to understand, and a proportionally more visible gain. A restaurant, a tradesperson, or a three-person agency feels the effect of an hour saved per day immediately. A large organization, by contrast, sees that same gain get lost in its fixed costs.

Will AI replace my employees?

The tasks that get automated are usually the ones nobody enjoys doing: data entry, follow-ups, first-pass sorting. Freeing that time up means giving more room to the work that creates value — and that AI can't do: customer relationships, creativity, judgment. The question isn't about replacing people, it's about replacing pointless tasks.

What budget should I plan to get started?

With the price cuts announced in late July 2026, a first agent handling simple tasks can end up costing just a few euros a month in real use. The main cost isn't the tool — it's the time spent scoping it out at the start: defining the task, the guardrails, and the validation process. That scoping work is what separates a profitable automation from a gimmick.

What if I'm not technical?

You don't need to be. Recent platforms offer visual interfaces where you plug in building blocks: receive an email, call a model, update a spreadsheet. The real skill involved is business scoping: knowing which task to automate, in what order, with what limits. And that's your job as a business owner, not a developer's.

Multi-tier AI, an opportunity for the region

The launch of GPT-5.6 marks a turning point: AI is no longer a luxury reserved for companies that can afford the most expensive models. It's becoming a modular service, where every small business can pick the right tier at the right price. For businesses in Anglet, Bayonne and Biarritz, this is a chance to get ahead in markets where most competitors haven't yet taken the leap.

What matters isn't being the first to automate everything. It's automating well: the right task, the right model, the right guardrail. And keeping people at the center of the decision.