OpenAI Opens GPT-5.6 Sol, Terra, and Luna to the General Public

OpenAI rarely makes noise without reason. When the company released public access to GPT-5.6 Sol — its highest-capacity model — on Thursday (July 9), while simultaneously launching the Terra and Luna versions, it did not merely expand a portfolio. It changed the cost-benefit calculation for every company that was still waiting for the right moment to take AI seriously.

Until then, Sol had been restricted to selected partners. The barrier was not only technical — it was also a signal that OpenAI wanted to control the pace of adoption of a model with advanced reasoning capabilities and autonomous agency. Now, with the opening, that containment is over.

What Sol, Terra, and Luna Are — and Why It Matters to Tell Them Apart

The three models form a tiered family, designed to cover different use cases and budget ranges. These are not "lite" versions with degraded quality, but specializations with distinct purposes.

GPT-5.6 Sol is the top of the line. It brings what OpenAI calls multi-step reasoning — the ability to decompose complex problems, plan sequences of actions, and execute them with minimal human supervision. It is the model closest to what the industry calls "autonomous agency": it does not merely answer questions, but acts within chained workflows. For a company, this means Sol can conduct a data analysis, identify anomalies, formulate hypotheses, and draft an executive report — without an analyst needing to intervene at every step.

GPT-5.6 Terra occupies the middle of the portfolio. At a significantly lower cost than Sol, it is optimized for structured processing tasks: document classification, information extraction, support ticket triage, contract summarization. Its quality-to-price ratio makes it the natural candidate for automations that need to run at high volume — thousands of interactions per day without compromising margins.

GPT-5.6 Luna is the entry layer. High speed, low latency, minimal cost. Recommended for FAQ chatbots, first-contact automatic responses, and any application where replies need to be fast and context is limited.

Why the Timing Matters for Brazil

The Brazilian corporate AI market is still, largely, in the proof-of-concept phase. According to data from Sebrae and FGV, more than 60% of Brazilian SMBs that experimented with some AI tool in the past two years did not advance to production deployment. The main obstacle reported is not lack of interest — it is infrastructure cost and uncertainty about returns.

The Sol + Terra + Luna logic directly attacks this problem. A company can run Luna for day-to-day operational volume, scale up to Terra for analytical tasks, and engage Sol only for the most critical or complex cases. This is not theory: it is a scalable cost architecture that makes it viable to start small without committing to expensive infrastructure from day one.

What Changes in Operations for Early Adopters

At FM Solutions & Consulting, we work with SMBs operating in very different contexts — from retail in São Paulo to agribusiness exporters. The pattern I see repeatedly is this: companies underestimate the cost of not automating and overestimate the cost of getting started.

With Sol's opening to the public, two workflows become immediately viable without major upfront investments:

1. Customer support with contextual reasoning. Unlike traditional rule-based chatbots or smaller models, Sol can handle ambiguous queries, cross-reference information from multiple internal sources (orders, history, policies), and make routing decisions with justification. The practical result: a reduction in unnecessary escalations to human agents and first-contact resolution rates above 70% — a figure we have observed in pilots with clients in the services sector.

2. Operational data analysis without a dedicated data scientist. SMBs rarely have an in-house data team. Sol, when fed with sales, inventory, or logistics spreadsheets, can identify seasonal patterns, flag deviations, and generate actionable recommendations in natural language. It does not replace deep statistical analysis, but it delivers 80% of the value in 20% of the time — and without hiring costs.

What to Evaluate Before You Start Testing

Open access does not mean every implementation will work. Three points deserve attention before starting a pilot:

• Quality of input data. Sol reasons well about what it receives. If a company's internal data is disorganized, the model will produce equally confusing analyses. Before the pilot, a data hygiene effort is worthwhile.

• Clear scope definition. AI projects that fail tend to have poorly defined scope. Start with a single workflow, measure the result, and then expand.

• Output governance. Especially in regulated contexts — legal, financial, healthcare — the model's output needs human review before generating action. Autonomous agency does not mean absence of accountability.

What Comes Next

The simultaneous opening of three models at tiered prices is a clear bet by OpenAI on the horizontalization of the market. The company is signaling that it wants to be present not only in large corporations with robust engineering teams, but inside any business with an internet connection and a real problem to solve.

For Brazilian SMBs, the window of competitive advantage is open. Those who start pilots now, still in 2025, will have six to twelve months of hands-on learning before the competition realizes it has fallen behind. That interval rarely repeats itself.