Free ChatGPT accounts are about to lose their message ceiling on text. That is the marquee item OpenAI announced today, and the timing is notable — the company says the product just passed 1 billion weekly users.
There is a qualifier, and it arrives quickly. The “unlimited” part covers text, and nothing else.
According to OpenAI, files, images, voice and image generation all remain governed by their own separate caps. In other words, an account that can now chat indefinitely will still hit a wall the moment you hand it a PDF or ask for a picture.
Powering the revamped free experience is GPT-5.6 Luna, now the default model for Free and Go users, taking over from GPT-5.5.
The Think button is the more interesting change
Free and Go tiers are also picking up a new “Think” button that turns up the model’s reasoning power for tougher questions. That amounts to reasoning-on-demand for people paying nothing — and that sort of capability has typically been the reason to upgrade in the first place.
Subscribers get a finer-grained take on the same concept. Plus and Pro users are receiving a thinking slider that adjusts “how much” thought goes into a response, calibrated to how complex a query is and how many steps it takes to resolve.
A button for the free crowd, a slider for the paying one. The hierarchy is behaving precisely as you would predict.
What Plus and Pro subscribers actually get
On the subscription side, Plus and Pro accounts unlock an upgraded GPT-5.6 Sol, which OpenAI frames as the stronger option for fast work: questions, web research, advice, planning, writing and decision-making. The company said the new model will deliver more compact answers.
One point is easy to overlook thanks to the naming. This is a distinct version from the GPT-5.6 Sol that powers Codex and Work, which stays as it is.
The error numbers OpenAI is putting forward
OpenAI said that in an internal evaluation, factual errors showed up 62% less often with GPT-5.6 Luna and 68% less often with GPT-5.6 Sol when stacked against GPT-5.5-Instant.
The key words there are internal evaluation. The figures come from OpenAI, benchmarked against OpenAI’s own earlier model, using a test OpenAI has not described. Making fewer factual errors than GPT-5.5-Instant is a different claim from making few factual errors.
Even so, a 62% drop is exactly the sort of number that gets stress-tested in a hurry once a billion people a week are using it.
When you get it
Plus and Pro users can use the updated GPT-5.6 Sol starting today. The remaining Free and Go changes roll out over the course of this week.
Unlimited text chats and the Think button for harder queries follow next week. Free-tier users should mark that one, and check the app before assuming the cap has actually lifted.