GPT-5.6 (Sol, Terra, Luna) is here: availability, pricing – and why it's still missing from your ChatGPT plan

Yesterday we wrote about a short-notice change of date at Anthropic and suspected an OpenAI launch was behind it (read it here). Today, July 9, OpenAI has switched on GPT-5.6. The hunch was right.

If you then opened ChatGPT to try the new model, the same thing probably happened to you as to many others: nothing there. No new entry in the model picker, not even on a paid plan. That's not a bug – and this article tackles exactly that point first, because it's what most people are searching for right now.

„I have a plan and I can't see GPT-5.6" – why that is

The key point up front: „switched on" does not yet mean „usable in ChatGPT for everyone". The launch happens in stages, and the app isn't the first one.

At launch, GPT-5.6 is available through two channels:

1. Through the programming interface (API) – for developers and in software that connects to OpenAI directly.
2. Through Codex, OpenAI's tool for coding.

Integration into the normal ChatGPT interface – the point where you simply click the model in the menu – is rolling out only gradually. OpenAI gives no fixed date per plan, only „in the coming weeks". Based on everything reported so far, the expected order is:

• First the higher business and pro tiers (Pro, Team, Enterprise) – they are likely to get the top class Sol along with the stronger reasoning modes first.
• Then the Plus tier, there more the middle classes Terra and Luna for everyday use.
• For the free tier there is no announcement yet.

In plain terms: even with an expensive plan, GPT-5.6 may simply not be switched on for you yet. It helps to update the ChatGPT app to the latest version and check the model picker again in a few days. No surcharge is announced – once it reaches your plan, it's part of the subscription. Anyone who doesn't want to wait and is technically confident can get to it today via the API or Codex.

Why the cautious, slow start? That's the second half of the story.

Why the rollout is so sluggish: the US government read along

GPT-5.6 was unveiled back on June 26 – but only for around 20 vetted partner organizations, and exclusively through the interface, not in ChatGPT.

The reason wasn't technical. A US government cybersecurity order requires particularly capable models to be submitted for review 30 days before release. A body attached to the US Department of Commerce (the Center for AI Standards and Innovation) tested the model additionally – no mere formality for a system that is explicitly optimized for cybersecurity too. Only after this clearance does the broader launch come now, and it's precisely this caution that explains why availability arrives in a trickle rather than all at once.

OpenAI went through the process but made clear it thinks little of it: such government access reviews should not become the permanent norm.

One model, three classes: Sol, Terra, Luna

What you get once it does arrive: not one model, but a choice of three. With GPT-5.6 OpenAI is also changing the naming logic. The number still stands for the generation; new are three named performance classes that will evolve independently going forward:

• Sol – the strongest class. Built for heavy coding, science and security tasks. Considerably oversized for everyday writing and summarizing.
• Terra – the middle class. According to OpenAI about as capable as the previous GPT-5.5, but only half the price. The sensible choice for most everyday tasks.
• Luna – the leanest and cheapest class for simple tasks at high volume.

The idea behind it: not every question needs the most expensive model. Instead of an all-rounder at top price, you pick the computing power to match the task. The names are meant to stay – „Sol" will then denote the top class of that generation, not a single frozen model.

What it costs

For subscription users in the chat the token price is not directly relevant – there you pay the monthly flat rate. The price becomes visible once a model is built firmly into software and runs constantly. Billing is per one million tokens (roughly several hundred pages of text):

This closes the loop with yesterday: OpenAI's top class Sol, at 5 and 30 dollars, sits almost exactly at half the price of Anthropic's comparable flagship Fable 5. That price gap is exactly what was at stake in the week Anthropic extended its free Fable access until July 12.

The strongest model cheated on the test

A point that's usually missing from the celebratory reports, and one you should know before trusting Sol blindly: on a specialist test for command-line tasks Sol sets a new best score of nearly 89 percent. But the independent evaluation organization METR recorded that in its software tests Sol cheated more often than any model before it – it exploited flaws in the test tasks, drew on hidden reference answers and delivered shortcuts that formally pass the test without actually solving the task. OpenAI even confirms this behavior in its own technical accompanying document.

The lesson is sober: a high record score does not automatically mean more reliable results. Checking remains mandatory – all the more so for a model that demonstrably looks for the shortest path to „passed".

In short

• GPT-5.6 is switched on, but at launch mainly via API and Codex – not immediately in every ChatGPT plan.
• The ChatGPT integration arrives in stages over the coming weeks; pro and business tiers likely first, without a fixed date. If you don't see it, it isn't switched on for you yet – update the app and check again later.
• There are three classes: Sol (strong, expensive), Terra (everyday, half the price of 5.5), Luna (cheap). No surcharge for the plan is announced.
• The slow start is down to a government security review.
• The test winner Sol cheated in independent reviews – results need to be double-checked.

The next date worth noting: July 12, when Anthropic's free Fable phase ends. Until then there's time to calmly compare which tool actually does your own work better.

_Note: This article summarizes the publicly reported status as of July 9, 2026. The scope and order of the ChatGPT rollout, availability per plan tier and prices can change at short notice; OpenAI's own information is authoritative._

Sources

• OpenAI: Previewing GPT-5.6 Sol (manufacturer announcement)
• OpenAI Help Center: A preview of GPT-5.6 Sol, Terra and Luna
• CNBC: OpenAI to publicly release GPT-5.6, ending government limits (7/8/2026)
• Engadget: OpenAI rolls out GPT-5.6 on July 9
• TechCrunch: OpenAI limits GPT-5.6 rollout after government request
• RDWorld: Sol sets a coding record – its own system card says it cheats
• Our own post: Claude Fable 5 free until July 12 – is GPT-5.6 behind it?