On 26 June 2026, OpenAI unveiled the GPT-5.6 family — and for the first time it didn't launch a single model but three: Sol, Terra and Luna. The change isn't just in the name but in the philosophy: instead of one giant model trying to be everything, there are now three tiers, each designed for a different balance of intelligence, speed and cost. For a business owner, that means a new decision: which model do you use for which task?

A new naming system: the number is the generation, the name is the tier

OpenAI explained that the number (5.6) identifies the generation, while Sol, Terra and Luna are durable capability tiers that each advance at their own pace. The names are inspired by the cosmos (sun, earth, moon), and the goal is to give users a clearer choice across three axes: intelligence, speed and cost. It moves away from the earlier "mini" and "nano" labels that confused more than they clarified.

The three differences at a glance

  • Sol (the peak): for the hardest work — complex coding and long sessions, advanced agents, and cybersecurity research. Highest capability and highest price.
  • Terra (the balanced): for large-scale business work — customer support, internal tools, document analysis. Strong performance at a lower cost, and the likely default for most companies.
  • Luna (the fast): for light everyday tasks — summarization, drafting and routine automation. The fastest and cheapest, performing near GPT-5.5 on several tests despite its low price.

Pricing: capability vs. cost

The clearest difference is price, per million tokens (input / output):

  • Sol: $5 input / $30 output — the same as GPT-5.5.
  • Terra: $2.50 input / $15 output — about half the cost of Sol.
  • Luna: $1 input / $6 output — the lowest cost.

In practice, the gap between Sol and Luna is large (up to five times on output), so choosing the right tier for each task translates directly into a very different monthly bill.

Why this launch is different

OpenAI didn't make the models available to everyone at once; it began with a limited preview to around 20 organizations, coordinated with the U.S. government following an executive order issued on 2 June 2026 on assessing new models' capabilities and safety before wide release. A general release is expected "in the coming weeks." It reflects a growing trend toward regulating the most powerful models before they ship.

What this means for your business

The principle we build on at Origami holds steady despite how fast models move: don't tie your system to a single model — let it pick the right tier for each task. In practice with GPT-5.6:

  • Start with Terra as the default for most business tasks — strong performance at a reasonable cost.
  • Route simple, high-volume tasks (summaries, replies, automation) to Luna to cut the bill.
  • Reserve Sol for tasks that truly deserve top accuracy — complex coding or long-running agents.
  • Always test on your real tasks and measure quality and cost together before committing.

This flexibility — a layer that lets you swap models and tiers without rebuilding the system — is what we design for our clients, because it protects your investment from a volatile model market and ensures you always pay the lowest possible cost for the quality you need.

Conclusion

The GPT-5.6 family doesn't ask "what's the most powerful model?" but "which tier for which task?". The winner is whoever builds a flexible system that distributes work intelligently across Sol, Terra and Luna — getting quality where it's needed and saving where the peak isn't.

Sources: OpenAI's official GPT-5.6 announcement (openai.com) and VentureBeat's report dated 26 June 2026. Pricing and details as stated in these sources.