GPT-5.6 is here: which ChatGPT model do you actually get now?

If you use ChatGPT, you probably saw that OpenAI launched GPT-5.6 and wondered about the only thing that really matters: does this change the chatbot you open every day? The short answer is yes, gently. On 9 July 2026 OpenAI made GPT-5.6 available to everyone, and if you pay for ChatGPT, the model behind your conversations is quietly switching over. Here is what that means in practice, with nothing left unexplained.

GPT-5.6 is not one model — it is a small family with three members, each tuned for a different job. Sol is the big, frontier brain for the hardest problems. Terra is the new all-rounder: OpenAI says it delivers roughly the same intelligence as last year's GPT-5.5 for about half the cost. Luna is the small, fast, cheap one for quick tasks. The switch that affects most people is simple: Terra becomes the default for paid ChatGPT — that is Plus, Team and Enterprise. If you are on the free tier, nothing changes yet; you stay on GPT-5.5. The rollout was staged over roughly a day or two, so it may have arrived quietly rather than with a banner in your app.

Here is the whole family at a glance, so you can see who gets what:

For developers building on the API, OpenAI published prices per million tokens: Sol is $5 in / $30 out, Terra $2.50 / $15 and Luna $1 / $6, and the gpt-5.6 name points to Sol. Each one can hold about a million tokens of context (a very long conversation or document) and write up to 128,000 tokens back.

Sol Ultra is best understood as a "think harder" setting. Instead of answering straight away, it takes more time, works through the problem in more depth, and can hand pieces of the job to smaller helper models before giving you an answer. You would reach for it on a genuinely tricky task — knotty code, a dense analysis — and skip it for everyday questions where the normal model is already quick and capable. OpenAI leans hard on efficiency here: it calls GPT-5.6 its strongest cybersecurity model yet and says it reaches that level using far fewer tokens, with Sam Altman claiming Sol is about 54% more token-efficient on coding.

Two other launches landed the same day. ChatGPT Work is a companion built for the office — it is designed to help with documents, spreadsheets and slide decks rather than just chat, so it fits into the actual files you work in. GPT-Live is a set of new voice models that can listen and speak at the same time, which is the technical way of saying conversations feel less like walkie-talkie turn-taking and more like talking to a person who can hear you even while they are answering. If you have ever been frustrated waiting for a voice assistant to finish before you could jump in, that is the friction this is meant to remove.

Here is the part that made this more than a normal product launch. GPT-5.6 did not appear the moment it was ready. Starting 26 June, OpenAI ran a 13-day "preview" limited to about 20 vetted partner organisations, because the model first went through a voluntary government review. That review comes from a June AI executive order under the Trump administration that asks makers of the most powerful "frontier" models to let the government look before a public release. In plainer terms: the tool you can now use was held back for almost two weeks so officials could check it first.

It is worth keeping this in perspective. OpenAI rates all three models "High" on cybersecurity and bio/chem risk under its own safety framework, but none reach the top "Critical" level — meaning, in its assessment, they cannot run a full cyberattack or build a working exploit on their own without a human driving. For you, the takeaway is small and practical: if you pay for ChatGPT, you are now getting Terra by default — the same intelligence you are used to, cheaper to run — with Sol and its Ultra mode there for the heavy lifting, and free users unaffected for now. Nothing to install, nothing to switch on; the upgrade already happened while the grown-ups argued about who gets to see it first.