ChatGPT Free is losing its message cap, so you can type as much as you want without hitting a wall. The catch is that those unlimited chats run on Luna, the lighter and less capable of the two models the company offers.
The Gist
- Free and Go accounts move to unlimited text chats, with GPT-5.6 Luna as the default model.
- Paying Plus and Pro accounts keep the stronger Sol model and gain a slider that controls how hard it thinks.
- Both models got more accurate: 62% fewer factual errors on Luna, 68% fewer on Sol.
- Image generation and file upload limits do not change.
Have ChatGPT Recap This Article
ChatGPTMeet Luna, the model your free account now runs on
OpenAI ships several versions of the same underlying AI, and they are not equally strong. Luna is the lighter one, built to answer fast and cheaply, while Sol is the heavier one that takes longer and handles harder questions better.
Until now, ChatGPT Free gave you a taste of the strong model and then stopped you. From this week, they get as many text messages as they want, but on Luna rather than Sol.
Go accounts, the cheaper paid tier, land in the same place as free ones on this change. Plus and Pro subscribers keep Sol, and they are the ones getting the upgraded version announced alongside.
If you are trying to work out whether paying is worth it for how you actually use the tool, we walked through that question in detail in our comparison of what the paid plan really adds over the free one.
How unlimited actually works here
Unlimited applies to typing, and nothing else. You can hold a conversation as long as you like, ask follow-up questions all afternoon, and nobody counts your messages.
The rest of the app keeps its ceilings. Uploading files and generating images stay capped exactly as before, which is where most free users bumped into limits anyway.
There is also a new Think button coming to free and Go accounts. It lets you tell the lighter model to slow down and put more effort into a specific question, rather than answering at its usual pace.
The rollout is staggered. The Sol improvements landed straight away, Luna becomes the default later in the week, and the Think button plus the unlimited chats arrive the week after.
Staggered launches are the norm here, and we saw the same approach when access to the newest model was handed out in stages rather than switched on for everyone at once. Your account may lag a few days behind a friend’s.
One more boundary is worth knowing. This whole update covers the regular ChatGPT app only, and the versions used at work or by developers are not affected by it.
Keep learning on AI Noobies:
- Google Assistant Leaves Your Phone in September
- Reward Hacking: Why AI Agents Cheat to Win
- New EU AI Rules Make Chatbots Say They’re AI
What a 62% drop in factual errors really means
Both models got an accuracy pass, and the numbers OpenAI published come from its own internal testing. Luna produces 62% fewer answers containing factual errors than the older GPT-5.5 Instant, and Sol produces 68% fewer.
Read that carefully, because it is a comparison, not a score. It says the new versions get things wrong less often than an older version did. It does not say they are right most of the time.
That distinction matters for anyone leaning on ChatGPT for facts. A big improvement over a shaky starting point is still not a guarantee, and checking anything that matters remains part of the job.
The paying side gets a second addition: a reasoning depth slider with five settings, available on web, mobile and desktop. It is a dial that lets you trade speed for care, question by question.
How to spot the change in your own account this week
Open ChatGPT Free and look at the model name near the top of the conversation. If it reads Luna, you are on the new default, and you can keep typing without watching a counter.
A good first test is a long back-and-forth on something low-stakes: planning a trip, drafting a message, working through a recipe. That kind of exchange is where an unlimited allowance genuinely changes the experience, because those conversations are exactly the ones that used to die halfway through.
When you hit something that needs real care, that is the moment to press Think once it reaches your account. Treat it as the button for anything you would be annoyed to get wrong.
Zooming out, this repeats the pattern set when GPT-5.6 first rolled out worldwide. Free users get the volume, paying users get the depth, and the gap between the two tiers becomes about quality rather than quantity.
Stay tuned on AI Noobies.
About the Author
