ChatGPT dropped the message limit for free users
On 6 August 2026 OpenAI announced that GPT-5.6 Luna becomes the default model for the Free and Go tiers, and that users on those tiers get text chat with no rate limits. The model switch landed that same week, and the removal of the message limit rolled out in the week of 10 August, along with a Think button that turns on deeper reasoning for a single question.
In practice a freelancer who had been saving messages for the questions that mattered can stop counting. Drafting a client email, summarising a call, wording a clause in a quote, checking a draft, all of that sits inside the free text that just opened up.
What stays capped
Text is the part that opened, not everything else. File uploads, image generation and the other tools inside ChatGPT still run against ceilings, and the chat itself is subject to abuse guardrails. Anyone planning to throw dozens of receipt and invoice PDFs at it at month end still hits a wall.
The free model is the smallest of three
Luna is the smallest of the three GPT-5.6 tiers, the fastest and cheapest in the family. Plus and Pro users get Sol, described as more reliable with facts, plus a slider that sets how much thought the model puts into an answer.
That difference matters exactly where a freelancer in Israel is tempted to lean on the chat: the VAT rate, tax brackets, whether an expense is deductible, the wording of a contract clause. A fast, cheap model gets those details wrong with full confidence. Free and unlimited is not the same as correct, and a number that comes out of the chat needs checking against the official source before it goes onto an invoice or a return.
What you type is training the next model
The "Improve the model for everyone" setting is on by default on personal Free, Plus and Pro accounts. Conversations are used to train future models unless you turn it off by hand, under Settings and then Data Controls. Turning it off applies to new conversations only. Data already included in a completed training run cannot be pulled back out.
For a one-person business this is the point that needs a decision, because what goes into the chat is rarely theoretical: client names, agreed prices, payment terms, sometimes the full text of an agreement. Removing the message limit only increases how much of that ends up there.
Where to draw the line
The simple rule is to separate wording from data. Wording, translating, shortening and tidying text is what a fast model does well, and if that content leaked it gives little away. Identifying client details, figures out of your books and the text of a signed contract sit on the other side, and there it is worth switching training off before pasting.
For anyone who works this way daily, switching it off is a one-time, thirty-second job, and it removes the need to weigh the question again on every conversation.