GPT-5.6 Sol, Terra, and Luna: What the Three-Tier Architecture Means for Your Business

Article spoiler:On July 9, OpenAI released Sol, Terra, and Luna as a single model family with three distinct capability and cost tiers.…We care about our clients, so we made a short takeaway from this article. Press to quickly get the point.

On July 9, OpenAI released Sol, Terra, and Luna as a single model family with three distinct capability and cost tiers. For SMBs that have been applying one model uniformly across all task types, this architecture formalizes a choice that previously had to be made manually and opens a practical path to building AI systems around task classes rather than around a single tool.

On July 9, OpenAI released GPT-5.6 following a short period of limited regulatory preview, and the structure of the release is as significant as the capability jump. Rather than a single flagship model, the company launched three tiers under one family name: Sol, Terra, and Luna. This architecture changes the primary question for businesses integrating AI from "which model should we use" to "which model should handle which class of task, at what cost, and with what level of risk tolerance."

For small and medium businesses, that shift in question creates both a more complex decision and a more useful one.

What Sol, Terra, and Luna actually do

OpenAI positions Sol as the flagship tier for the most demanding professional tasks: deep strategic analysis, complex software development, long multi-step reasoning chains, and high-stakes decisions where the cost of an error is material. Sol is the most capable and the most expensive.

Terra functions as the balanced working layer, designed for the volume of everyday business work: long-form documents, internal analysis, content generation, and strategic planning. Its capabilities sit close to GPT-5.5, and it is priced at roughly half the cost of Sol for the same task.

Luna is optimised for speed and volume: short responses, draft generation, classification, data extraction, and bulk processing. It is the fastest and the cheapest tier in the family.

All three models share a context window of approximately 1.05 million tokens and support up to 128,000 output tokens, making it possible to work with large document sets, full CRM exports, or long operational logs in a single request.

The pricing structure reflects the tier differentiation clearly:

The five-to-one difference in output token cost between Sol and Luna makes tier selection a budget decision that compounds quickly at scale.

The architecture shift this creates for SMBs

Most SMB AI workflows have been built around a single model applied uniformly across all task types, from customer support drafts to strategic analysis. The GPT-5.6 family formalizes a principle that was previously a manual choice: different task classes carry different capability and cost profiles, and building a routing layer between tiers produces better economics than selecting a single tier and using it everywhere.

The Sol/Terra/Luna structure offers something like a ready-made routing framework: one brand, three levels of compute and reasoning, with pricing that reflects the actual value differential between task types.

For SMBs, the practical implication is the opportunity to design AI systems around task classes and margin requirements rather than around a single model choice. Strategic tasks go to Sol, operational volume goes to Terra, and bulk routine work goes to Luna. That approach reduces token costs in high-volume workflows while preserving access to flagship capability for the tasks where it has a measurable impact on quality or risk.

How GLC thinks about this for clients

As of July 9, our recommendations for SMBs working with or considering GPT-5.6 fall into four areas.

Terra as the working baseline. For most SMBs, Terra is the more appropriate starting point for everyday processes: it delivers quality close to GPT-5.5 at substantially better economics and was designed specifically as the operational workhorse of the family. Building workflows around Terra by default, with Sol reserved for genuinely high-stakes tasks, keeps costs manageable at scale.

A two- or three-tier architecture rather than a uniform choice. Terra handles internal processes, from reports to content generation. Luna covers high-volume support and data processing. Sol enters as a separate layer for strategic and high-risk decisions: architecture design, complex analytics, critical code. The value of this structure increases proportionally with the volume of lower-tier tasks.

Token economics before launch. With a five-to-one difference in output token cost between Sol and Luna, selecting the wrong tier for a volume workflow burns budget in ways that are invisible until the invoice arrives. Modelling the load before deployment — how many requests will flow through Luna, how many through Terra, and which specific scenarios genuinely require Sol — is faster to do on a spreadsheet than to correct in production.

Sol as a laboratory, Terra and Luna as production. Experimenting with complex new logic, new product concepts, and multi-step chains in Sol, then migrating proven patterns to Terra or Luna for scale, gives you the full capability of the flagship tier during design and the economics of the lower tiers during operation.

Specific scenarios by business area

Sales and marketing. Strategic positioning work, messaging frameworks for new markets, and the reasoning behind campaign architecture belong in Sol, where the depth of analysis reduces the risk of a poor strategic bet. Regular content production — newsletters, landing pages, social posts — runs efficiently on Terra. Bulk A/B testing of headlines, descriptions, and ad copy scales cleanly on Luna, where thousands of variants can be generated and filtered at minimal cost.

Customer support and operations. First-line support automation — FAQ responses, request routing, quick answers — runs on Luna, where speed and cost per interaction matter most and the tasks are sufficiently defined. Complex support cases that combine documentation, customer history, and business logic belong on Terra as the primary layer, with escalation pathways to Sol for rare critical situations.

Internal analytics and management decisions. Analysing large internal datasets — reports, CRM exports, correspondence, log files — is well-suited to Terra, which balances context capacity, output quality, and cost for the typical SMB analytical workload. Multi-step scenarios from data to decision, such as market analysis with scenario modelling and recommendation generation, belong in Sol as the reasoning layer of a more complex agent process.

ChatGPT Work: the agent layer on top

Alongside the GPT-5.6 launch, OpenAI introduced ChatGPT Work, positioned not as a model but as an execution layer: a tool capable of operating across your applications and files, running a project over hours, and converting a goal into completed work rather than a response.

The agent combines the capabilities of Codex for development tasks and ChatGPT for dialogue and logic, with Sol, Terra, and Luna as the underlying models. For SMBs, this means that model selection extends beyond "which model answers my question" and becomes part of a workflow design: which class of task goes to the agent, at which tier it runs, and where human review enters the process.

Our approach at GLC is to help clients describe their business processes as "jobs to be done" that an agent can execute, then optimise the Sol/Terra/Luna combination for each workflow against quality, cost, and control requirements.

What to do before the end of July

Five actions that translate this release into a practical change in how your AI systems work:

1. Map the business processes where you currently use AI, and where AI could have the highest impact, across sales, support, operations, and analytics.
2. Classify those processes by task type: strategic, operational, or bulk. That classification becomes the routing map for Sol, Terra, and Luna.
3. Run one pilot using a Terra plus Luna routing setup for a real workflow, measure costs and output quality, and use that as the baseline for further decisions.
4. For high-risk, high-value tasks — significant deals, legally sensitive decisions, IT architecture choices — plan separate Sol experiments with mandatory human review before any output is acted on.
5. Assign ownership of AI architecture inside your team: a person or a partner who looks at model economics, process fit, and risk, rather than just at model features.

The Sol/Terra/Luna launch is, in our reading, less an incremental model update and more a structural invitation to move from a single-tool approach to a deliberate, layered AI architecture built into real business processes.

If you want to map your current workflows against this framework and identify where the architecture applies to your specific situation, a first conversation covers the ground in under an hour. Get in touch.