What Is GPT-5.6? OpenAI’s Sol, Terra & Luna, Explained

OpenAI just released GPT-5.6, and if the name alone left you with questions, you are not the only one. It is not one model, it is three. It shipped with a brand new naming system. And it landed in the middle of the same government review that briefly took Claude offline last month.

Here is the plain-English version: what GPT-5.6 actually is, what Sol, Terra, and Luna mean, what it costs, and whether any of it changes how you use AI day to day.

What is GPT-5.6?

GPT-5.6 is the latest generation of OpenAI’s general-purpose models, the technology behind ChatGPT. It went out first as a limited preview to a small group of trusted partners in late June 2026, then opened to the public on July 9. OpenAI says it is available across ChatGPT, its Codex coding tool, and the OpenAI API that other apps are built on.

The pitch is not one dramatic leap in intelligence. It is a family of models aimed at real work: enterprise tasks, coding, scientific research, and cybersecurity. And the story underneath the launch is less about raw power and more about doing the same work for less money.

Sol, Terra, and Luna: the names, explained

This is the part that trips people up, so here it is plainly. With GPT-5.6, OpenAI split the name into two pieces:

• The number (5.6) tells you the generation.
• The name (Sol, Terra, or Luna) tells you the capability tier. OpenAI says these tiers can now each improve on their own schedule, rather than all moving together.

So "GPT-5.6 Sol" is the top tier of the 5.6 generation. A useful way to picture it: same model year, different trim levels, like a car.

What actually changed

The clearest theme is efficiency. OpenAI CEO Sam Altman said Sol is "54% more token efficient when it comes to AI coding" compared with previous versions. In plain terms, tokens are the small units of text an AI reads and writes, and they are what you pay for, so using fewer of them for the same result means lower cost.

On coding specifically, OpenAI reported that Sol set a new state-of-the-art score of 80 on its coding evaluation, about 2.8 points ahead of Claude’s Fable 5, while using less than half the output tokens. That is OpenAI’s own claim, measured on its own test, so treat it the way you would treat any launch-day benchmark. Still, the direction is telling: the race is shifting from "biggest" to "most efficient."

Here is what the three tiers cost through the API, per million tokens:

OpenAI also introduced ChatGPT Work alongside the models. It is described as a workplace companion for teams that runs on desktop, web, and mobile and helps with everyday clerical tasks.

Why the government got involved

GPT-5.6 was supposed to arrive a little earlier. It was held back by a US government review, and this is where the story connects to one we covered last month.

In early June 2026, a White House AI cybersecurity order asked leading AI companies to voluntarily submit their most powerful models for government review 30 days before releasing them publicly. OpenAI staggered its rollout to comply. According to reporting, the Department of Commerce’s Center for AI Standards and Innovation ran extra testing and held more meetings with the company, OpenAI sent technical experts to Washington, and the review was ultimately sped up so the public launch could happen on July 9.

If that sounds familiar, it should. The same government push briefly took Anthropic’s Claude models offline in June, when a directive required blocking foreign-national access to Fable 5 and its research sibling Mythos 5. We wrote about that here: When will Claude Fable 5 be back? Two of the biggest AI labs, tangled in the same review, within weeks of each other. It is a clear sign of where AI oversight is heading in 2026.

What this means for you

Here is the honest answer most launch coverage skips: if you are a normal person and not an AI engineer, you probably do not need to memorize any of these names.

Model names change every few weeks now. GPT-5.5 became GPT-5.6. Claude shipped Fable 5. Next month there will be another. Chasing the single "best" model is a part-time job, and the gaps that matter to a real person, writing an email, planning a trip, fixing a bug, keep shrinking as every top model gets good enough.

What actually helps is a tool that makes the model choice for you and turns it into finished work. That is the whole idea behind agentic AI: instead of one chatbot and a menu of model names, you get expert agents that plan and complete real tasks, running on frontier models under the hood.

That is exactly how our app works. Agentic AI gives you a team of ready-made expert agents on your iPhone and iPad, running on frontier models, so you get the capability without the model-name homework. There is no account and no login, and you can put a whole team of agents on one task. When a model like GPT-5.6 raises the bar, the app you use is what turns that into something you can actually hold.

Quick answers

Is GPT-5.6 free? The models are available through ChatGPT and the paid OpenAI API. The prices above are the API rates. What you can access inside ChatGPT depends on your plan.

GPT-5.6 vs GPT-5.5? 5.6 is the newer generation. OpenAI positions Terra as competitive with GPT-5.5 at roughly half the cost, and Sol as a clear step up, especially for coding and efficiency.

Is it better than Claude? OpenAI says Sol edges out Claude Fable 5 on OpenAI’s own coding test. Independent testing takes time, and "better" depends on the task. For most people, the app you use matters more than the exact model inside it.

How do I try GPT-5.6? Through ChatGPT, OpenAI’s Codex, or the API. Or skip the model shopping and let an agent app use frontier models for you.

Sources: OpenAI: GPT-5.6 · TechCrunch · CNBC · Engadget · Axios