GPT-5.6 Explained (2026): Sol, Terra, Luna, and the Catch
OpenAI previewed GPT-5.6 on June 26, 2026, and the headline is not just the model. It is who is allowed to use it. According to OpenAI and reporting by Axios, GPT-5.6 launched under a limited release at the request of the U.S. government, with access restricted to around 20 approved companies at first. Here is what the model is, who can actually touch it, and the question that matters for anyone handling sensitive work.
This is a developing story. Treat availability and capability claims below as of late June 2026, and verify the current state on OpenAI's own pages before acting.
The three versions: Sol, Terra, Luna
OpenAI is releasing GPT-5.6 in three variants, per its announcement:
- Sol. The most powerful version.
- Terra. A balance of efficiency and power.
- Luna. Built for speed and affordability.
OpenAI also says it will add options for more reasoning, plus an "ultra" mode that splits work across multiple sub-agents.
What it is reportedly good at
OpenAI positions GPT-5.6 as a step up in coding, science, and cybersecurity, paired with what it calls its most advanced safety stack. The cybersecurity gains are exactly why the release is restricted: the concern is that a stronger model is also better at finding software vulnerabilities.
On that point OpenAI says it believes "GPT-5.6 Sol is better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks," and that the model's capabilities do not reach the "critical" level in its own preparedness framework.
The catch: you probably cannot use it yet
This is the part most coverage buries. At launch, GPT-5.6 is a limited preview for roughly 20 government-approved companies. OpenAI says it expects to expand access next week and aims for a broad release in the coming weeks, but at the time of writing it is not generally available.
OpenAI itself is not thrilled about the gate. In its blog post it said this kind of government access process should not become the long-term default, because "it keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them." The restriction follows similar U.S. moves on Anthropic's latest models, so this is becoming an industry pattern, not a one-off.
The practical takeaway: for almost everyone, GPT-5.6 is an announcement, not yet a tool you can put to work.
The question that does not change with the version number
New model, same question for anyone handling confidential work: where does your data go? Whichever GPT you use, on consumer plans the conversation can be retained and, depending on plan and settings, used to train the model. A more capable model does not fix that. For a patient detail, a client file, or unreleased work, you still need retention and no-training commitments in writing, plus EU processing if you are in Europe.
So the upgrade that matters for regulated and client-facing work is not Sol over Terra. It is private over default.
How to use frontier AI privately, today
You do not have to wait for a gated model, or send confidential work to a consumer product, to use frontier AI now.
The Wysor solution
Use frontier AI privately, in the EU, today.
While GPT-5.6 is gated to a handful of approved companies, you can run today's frontier models, GPT-5, Claude, and Gemini, in one private workspace, processed in the EU with zero retention.
Frontier models, available now
GPT-5, Claude, Gemini and more
Zero data retention
By contract, on every plan
EU-hosted, signed DPA
No training on your data
When GPT-5.6 opens up, the same privacy rules will decide whether you can use it for sensitive work. A workspace built for zero retention and EU processing is how you stay ready either way.
Frequently asked questions
What is GPT-5.6? GPT-5.6 is OpenAI's model previewed on June 26, 2026, in three variants: Sol (most powerful), Terra (balanced), and Luna (fast and affordable), with an "ultra" mode that splits work across sub-agents. OpenAI highlights gains in coding, science, and cybersecurity.
What is the difference between Sol, Terra, and Luna? Per OpenAI: Sol is the most powerful, Terra balances efficiency and power, and Luna is built for speed and affordability.
Can I use GPT-5.6 right now? For most people, no. At launch it is a limited preview for around 20 government-approved companies. OpenAI says it expects to widen access in the following weeks and aims for a broad release, but it is not generally available at the time of writing.
Why is GPT-5.6 access restricted? Because of its stronger cybersecurity capabilities. The U.S. government asked OpenAI to limit access while a review framework is developed. OpenAI says it does not consider this a sustainable long-term approach.
Is GPT-5.6 private? The same caveats apply as with any consumer AI: on consumer plans, conversations can be retained and used to train the model by default. For confidential or regulated work you need zero retention and EU processing, regardless of model version.
How can I use frontier AI privately while GPT-5.6 is gated? Wysor gives you GPT-5, Claude, and Gemini in one private workspace, processed in the EU with zero data retention and no training on your data.
Keep reading
- Claude vs ChatGPT: Which Is Better, and Why Not Both?. The comparison everyone asks about.
- Is ChatGPT Safe? What It Does With Your Data. The privacy question in detail.
- Sovereign AI: What It Means and How to Get a Private LLM. Keeping control of your data.
Questions about using frontier models privately? Reach us at [email protected] or through the in-app contact form.
