GPT-5.6 Sol Explained: What's New and What It Means for You

The short version: GPT-5.6 is out. After a slow, government-coordinated preview, OpenAI launched the GPT-5.6 family for general availability on July 9, 2026 -- the flagship Sol, plus Terra (a balanced everyday model) and Luna (the fast, cheap one) -- across ChatGPT, Codex, and the API. If you pay for ChatGPT (Plus and up), you can now pick Sol in the model menu. If you are on Free or Go, you do not get Sol -- since August 6, 2026 you get Luna as your default instead. Your subscription price does not change. This page explains what GPT-5.6 does, who gets which model, and whether you should care.

What Changed (July 9)

GPT-5.6 moved from a restricted preview to general availability. When this page first published on July 4, none of these models were usable in ChatGPT -- they sat in a limited preview for vetted partners, coordinated with the US government because Sol is unusually strong at cybersecurity work. That preview is now over. As of July 9, 2026, GPT-5.6 is rolling out to everyone across ChatGPT, Codex, and the OpenAI API, "starting globally now" and reaching full availability over roughly the next 24 hours. The paragraphs below are updated to reflect that you can now use it.

Key Takeaways

• GPT-5.6 went generally available on July 9, 2026, across ChatGPT, Codex, and the API, after a government-coordinated limited preview.
• Three models: Sol (flagship), Terra (balanced, everyday), and Luna (fast and cheap). Sol, Terra, and Luna are OpenAI's new naming for capability tiers.
• Do not confuse these with GPT-6 Sol and GPT-6 Luna. OpenAI reused both names one generation up on September 22, 2026. The GPT-6 pair is not usable in ordinary chat at all: OpenAI says they "are not available in regular ChatGPT conversations." This page is about the GPT-5.6 models, which are the ones in your chat today. See GPT-6 Sol and Luna explained.
• Who gets Sol in ChatGPT: Plus, Pro, Business, and Enterprise users, through medium and higher effort settings. Pro and Enterprise can also pick Sol Pro for the hardest tasks.
• Free, Go and logged-out users do not get Sol. OpenAI now states it on two pages -- its model release notes, and its GPT-5.6 help page, rewritten August 9, 2026: "Free and Go users do not have access to GPT-5.6 Sol. Their default model is GPT-5.6 Luna, which also powers Think." Luna became that default from August 6, 2026. In ChatGPT Work and Codex it names Terra for those plans -- a different product, and unchanged by the August 6 release.
• Your plan price is unchanged. The per-token prices OpenAI published apply to developers on the API, not to ChatGPT subscribers.
• The settings have changed since launch. The July launch shipped a max reasoning effort and ultra mode. In ChatGPT those became a Thinking slider: Instant, Medium, High, Extra High and Pro. Plus gets Medium and High only.
• Since September 14, 2026, ChatGPT does not add thinking time by itself on Plus or Pro. OpenAI retired automatic switching from Instant to Thinking on those plans globally, and removed the Higher intelligence setting from ChatGPT on the web for them. It "can still switch automatically for safety purposes."
• OpenAI's own benchmarks put Sol ahead of Claude Fable 5 on several agentic and coding tests while using fewer tokens -- vendor-reported numbers, so treat them as claims, not neutral verdicts.

What Is GPT-5.6 Sol?

GPT-5.6 Sol is the flagship of OpenAI's new three-model family, and OpenAI calls it its strongest model yet. The family covers three price-performance tiers: Sol for the hardest reasoning and long, multi-step "agentic" work, Terra as the balanced model for everyday tasks, and Luna as the fast, affordable option. OpenAI describes Terra as delivering GPT-5.5-competitive performance at lower cost, which tells you the real story for most people: Sol is the headline, but the cheaper tiers are what most ChatGPT users will actually end up talking to, because Sol is reserved for paid plans and the paid default leans on the smaller models for routine questions.

If the sun-and-moons naming feels new, that is because it is. OpenAI has moved away from confusing suffixes like "mini" and "nano" toward names you can actually remember: Sol is the big one, Luna is the small one, and the number (5.6) just marks the generation.

That number is now the part to read carefully. On September 22, 2026 OpenAI shipped GPT-6 Sol and GPT-6 Luna, reusing both names a generation up. They are different models in different places: OpenAI says the GPT-6 pair "are not available in regular ChatGPT conversations" and live only in ChatGPT Work and Codex. Everything on this page is about the GPT-5.6 models, which are the ones running in the chat you open. If someone tells you they are using Sol, the follow-up question is which generation. See GPT-6 Sol and Luna explained.

What Can It Do That GPT-5.5 Cannot?

Two genuinely new settings, plus a broad capability jump. First, GPT-5.6 supports a new max reasoning effort -- a setting that gives the model more time to think through a problem than the previous "xhigh" setting allowed. Second, there is ultra mode, which OpenAI says coordinates four AI agents working in parallel by default to finish demanding jobs faster. If "agent" is a fuzzy word for you, our plain-terms explainer on what an AI agent actually is covers it in two minutes. Most everyday users will never touch either setting; they exist for developers and power users running heavy, long-running tasks.

On measurable skills, OpenAI reports state-of-the-art results across coding, agentic work, cybersecurity, and science, and says GPT-5.6 gets more useful work out of each token than rivals. On a long-horizon workflow benchmark called Agents' Last Exam, OpenAI puts Sol ahead of Anthropic's Claude Fable 5, and says Terra and Luna beat Fable 5 at a fraction of the cost. Read those as vendor-reported claims measured on OpenAI's chosen tests -- the direction is real, but the exact margins come from OpenAI, not an independent referee.

Why the Rollout Was So Slow

The preview was a policy story as much as a product story. Before launch, OpenAI discussed GPT-5.6's capabilities with the US government and, at the government's request, shared the list of organizations in the early-access program. The staggered rollout gave federal reviewers time to assess national-security risk from the model's offensive-security potential -- Sol is exceptionally good at finding software vulnerabilities -- before it reached the general public. OpenAI says GPT-5.6 launched "with our most robust safeguards to date" after its most extensive evaluation period yet.

This mirrors what happened to Anthropic weeks earlier, when its Claude Fable 5 and Mythos 5 models were briefly hit with export controls. Frontier AI releases in 2026 now routinely involve Washington, and staggered launches like this one are becoming the norm rather than the exception.

Do You Get It, and on Which Plan?

Only if you pay. Here is where things stand for ChatGPT, using OpenAI's own wording rather than the tidier version most coverage prints:

Sources: OpenAI's model release notes for the paid/free split, its GPT-5.6 launch post for the effort settings, its ChatGPT release notes of August 6, 2026 for Luna as the Free and Go default, and its GPT-5.6 in ChatGPT help page, rewritten August 10, 2026, which states it a second time, confirms that "Free and Go users have unlimited everyday text chats, subject to abuse-prevention safeguards," and adds that the free Think button "uses GPT-5.6 Luna, not GPT-5.6 Sol."

For developers, OpenAI published API prices per 1 million tokens: Sol is $5 input / $30 output, Terra is $2.50 / $15, and Luna is $1 / $6. Those figures are for API usage and do not change what ChatGPT subscribers pay each month. The rollout is gradual over about 24 hours, so if you do not see GPT-5.6 in your model picker yet, it is still reaching your account -- there is nothing to buy or switch on beyond your existing plan.

What Changed on September 14, 2026: No More Automatic Thinking

ChatGPT stopped deciding for you, on Plus and Pro. In its ChatGPT release notes dated September 14, 2026, OpenAI says: "We're retiring automatic switching from Instant to Thinking (reasoning) for ChatGPT Plus and Pro users globally," and "We're also removing the Higher intelligence setting from ChatGPT on the web for these plans." One exception stays: "ChatGPT can still switch automatically for safety purposes."

Its help page, updated September 16, 2026, closes the obvious workaround too: "Asking ChatGPT to 'think harder' or 'think deeply' won't automatically switch to a different thinking level." If you want more thinking, you pick it.

What you pick from is a slider with five positions, and your plan decides how far it goes:

Free and Go are not on the slider at all. OpenAI's line for them: "Free and Go users can use Think for harder questions. Think uses GPT-5.6 Luna, not GPT-5.6 Sol."

What this means in practice. If you are on Plus or Pro and you were used to ChatGPT quietly taking longer on a hard question, it will not any more. Short answers to hard questions are now your cue to move the slider yourself, and it is a genuine downgrade in convenience dressed as a control. If you are on Business, the old behaviour survives: OpenAI says Higher intelligence is still under Settings > General there, where it "controls whether ChatGPT automatically uses additional thinking while Instant is selected."

Should You Care?

Care a little, act only if you already pay. If you are on Free or Go, the flagship is still paid-only -- but the family did eventually reach you: since August 6, 2026 your default is GPT-5.6 Luna, and OpenAI has now confirmed it paired that with unlimited everyday text chats and a Think button, which reached the web for Free and Go on August 14, 2026 after launching in the mobile app. For routine questions you are unlikely to notice a dramatic difference between Luna and Sol either way. If you are on Plus or Pro, the practical win is that Sol is available in your model picker for harder tasks, and it is designed to reach good answers using fewer tokens and less time, which mostly matters when you hit usage limits.

Pick Sol when you have a genuinely hard, multi-step task -- deep research, complex coding, careful analysis. Stick with the default for everyday questions, where it is faster and the quality gap is small. Do not upgrade your plan just for this unless you were already weighing it; a new flagship in your existing plan is not a reason to spend more.

GPT-5.6 does not exist in a vacuum: Google's Gemini 3.5 Pro is in partner testing with no announced date, while its free Gemini 3.6 Flash landed on July 21, and Anthropic's Claude Fable 5 has been trading the capability crown with OpenAI all year. If you mostly want an everyday assistant, the practical differences are smaller than the headlines suggest -- see our ChatGPT vs. Gemini everyday-use comparison and our guide to which AI chatbot is worth paying for.

The bottom line: GPT-5.6 Sol is a real, significant release, and it is a paid one. Plus and up get the flagship in their model menu; Free and Go are excluded from Sol and run GPT-5.6 Luna instead, which OpenAI confirmed on August 6, 2026. Nobody's subscription price went up. If you pay, try Sol on something genuinely hard and let the default handle the rest.

Corrections and Updates

September 29, 2026: this page said ChatGPT Pro has two tiers. OpenAI's release notes of September 29 added Pro 500 at $500 a month. The plan table and the pricing FAQ above now say Pro starts at $100 and has three tiers. Nothing about GPT-5.6 access changed.

September 17, 2026: this page described "max" reasoning effort and "ultra" mode as the settings you get in ChatGPT. Those are the July launch's names, and OpenAI has replaced them in the chat app with a Thinking slider (Instant, Medium, High, Extra High, Pro), on which Plus reaches only Medium and High. It also retired automatic switching from Instant to Thinking on Plus and Pro globally on September 14, 2026 and removed the Higher intelligence setting from ChatGPT web on those plans. The summary, Key Takeaways and the ultra FAQ were corrected, and a new section documents the slider.

August 31, 2026: ChatGPT Pro is two prices, and this page printed only the higher one. (Superseded September 29, 2026: a third Pro tier, Pro 500 at $500 a month, was added.) It said "Pro is $200"; OpenAI has sold Pro at $100 and $200 a month since April 9, 2026, same capabilities, different usage allowances, so the entry price for Sol Pro is $100.

August 8, 2026: OpenAI has now named the free model. This page said OpenAI "does not name which GPT-5.6 model you do get" on Free and Go, which was true when written; its August 6, 2026 release note says GPT-5.6 Luna became the default for Free and Go.

August 4, 2026: this page said "Free and Go users get Terra" as a plain fact about the ChatGPT app. That was wrong: OpenAI's model release notes say "Free, Go, and logged-out users are not included," and the only place it names Terra for those plans is ChatGPT Work and Codex, a different product.