OpenAI Previews GPT-5.6 Family: Sol, Terra and Luna Launched in Limited Rollout

For most people, the name GPT-5.6 won’t mean much yet. But if you work in software development, cybersecurity, scientific research, or any field that relies heavily on AI tools, what OpenAI announced this week is worth paying close attention to. The company has unveiled a new family of three artificial intelligence models — Sol, Terra, and Luna — each built for a different kind of work, and each priced accordingly. The catch? You can’t sign up to try them. Not yet, anyway.

What Is GPT-5.6, and Why Three Models?

OpenAI’s decision to release three distinct models under the GPT-5.6 banner reflects a broader industry trend: one size no longer fits all when it comes to AI. Different tasks demand different trade-offs between speed, cost, and raw capability.

Sol is the flagship. OpenAI describes it as its most capable model to date for cybersecurity work and its go-to option for what the company calls “frontier reasoning and long-horizon work” — meaning complex, multi-step tasks that unfold over time rather than simple one-off questions. Think advanced code generation, deep research, or extended problem-solving chains.

Terra sits in the middle. It’s described as a balanced, lower-cost option for everyday professional work — the kind of model a developer or knowledge worker might reach for when they don’t need Sol’s full firepower but still want reliable, high-quality results.

Luna is the speed runner. Designed for high-volume tasks where cost efficiency matters most, it’s the model you’d deploy when you need to process enormous amounts of data quickly and cheaply.

The API pricing reflects those differences. Sol costs around £3.90 per million input tokens and around £23.50 per million output tokens. Terra comes in at around £1.95 input and around £11.75 output. Luna is the most affordable, at around £0.78 input and around £4.70 output. (These figures are converted from OpenAI’s published USD pricing at approximate current rates.)

A Limited Preview, Not a Public Launch

This is not a general release. OpenAI has been clear about that. The models are currently available only through the OpenAI API and Codex — its AI-powered coding platform — to a small group of trusted partners and organisations. ChatGPT users won’t see Sol, Terra, or Luna in their interface during this preview period. There is no public application process and no waitlist to join.

OpenAI says it plans to bring the full family to ChatGPT, Codex, and the broader API in the coming weeks, but no specific date has been confirmed.

The U.S. Government Connection

One detail that stands out in OpenAI’s own documentation is that this preview was started partly at the request of the U.S. Government. OpenAI says it engaged with U.S. Government bodies before launch, and that participation in the current preview is limited to trusted partners whose involvement has been shared with government. It’s an unusual disclosure, and it speaks to how seriously regulators and governments are now treating the deployment of frontier AI systems.

On the safety side, OpenAI has classified Sol, Terra, and Luna as “High capability” in cybersecurity and in biological and chemical risk under its internal Preparedness Framework — but not at the highest “Critical” level. That’s a meaningful distinction. The company also says that in testing, Sol and Terra were unable to carry out autonomous, end-to-end attacks against hardened cybersecurity targets. OpenAI describes the safeguards for this launch as its most sturdy yet.

The models do not reach OpenAI’s High threshold for AI self-improvement, the company adds — meaning they are not considered capable of much accelerating their own development.

A Technical Upgrade Worth Noting

Beyond the three new models themselves, GPT-5.6 introduces a more predictable approach to prompt caching. For those unfamiliar: prompt caching allows an AI system to store and reuse parts of a conversation or document rather than processing the same text from scratch every time, which reduces both cost and processing time. OpenAI is now offering explicit cache breakpoints and a minimum cache life of 30 minutes. Cache writes are billed at 1.25 times the standard uncached input rate, while cache reads receive a 90 per cent discount on cached input — a meaningful saving for developers running large-scale applications.

Sam Altman, OpenAI’s chief executive, has not made a specific public statement on GPT-5.6 beyond the company’s official announcements. But the documentation itself is unusually detailed, suggesting OpenAI is keen to demonstrate both the capability and the care that has gone into this release.

What This Means for Kent Residents

For most people here in Kent right now, the practical impact is limited — this preview is closed to the public, and no local organisations have been named as participants. But once OpenAI opens GPT-5.6 to general availability in the coming weeks, businesses, developers, and public sector digital teams across the county could find new tools at their disposal, chiefly if they already use the OpenAI API or Codex. Whether that translates into better services or lower costs will depend on how quickly organisations in Kent choose to adopt what’s on offer.