OpenAI has quietly removed the rate limit that defined the ChatGPT Free experience for the past two years. Free and Go tier users used to hit a wall after ten to forty text messages in a three to five hour window, after which the chat was pushed down to a smaller, weaker model until the window reset. That throttle is gone, and the chat does not degrade the way it used to. Free users now get unlimited text messages on OpenAI’s lightest current model, GPT-5.6 Luna, which replaces the older GPT-5.5 Instant as the default.
The change does not turn Free into Plus. The things that used to cost money still cost money, and a few of the new limits are not where most users expect them. If you came back to ChatGPT after a few months away, the right way to read the announcement is “Free is more useful for short text tasks; Free is still the wrong tier for anything else.”
A reasonable mental model is to think of Free now as a text-first chat surface that is generous on the text path and unchanged everywhere else. If your workflow fits inside that mental model, Free is now a real option. If your workflow needs any of the tool surfaces that still cost money, the announcement is not news for you.
What “unlimited text” actually means
The new limit applies to plain text messages. If a request involves file uploads, image attachments, image generation, or voice mode, the old rate limits still apply. The same is true for code interpreter sessions and any tool that runs against an attachment. The text-only path is the path that opened up; every other surface on Free is unchanged.
For a user who mostly pastes text questions and reads the answers, this is the most generous Free has ever been. For a user who uploads PDFs, generates images, or uses voice mode, the experience is essentially identical to last week, and the marketing language around “unlimited” is misleading for those use cases. The text-only path is the path that opened up; every other surface on Free is unchanged, and the rate limits on those surfaces are the same numbers that applied before the announcement.
Why the lightest model
The default model on Free and Go is now GPT-5.6 Luna, the smallest model in the GPT 5.6 family. Luna is described by OpenAI as faster, more reliable, and more focused than the Instant model it replaces. The trade-off is capability. Luna is not the model you reach for when a question needs long reasoning or careful citation. It is the model that handles short factual answers, summaries, and simple drafts quickly and without surprises.
If you want a stronger model on Free, the new “Think” button asks the model to spend more time before responding. The button is the only way to get a deeper answer on the Free tier; without it, Luna will give you the fastest answer it can, even when a slower answer would have been the better one.
For most of the questions Free users actually ask, Luna is a fine default. Short factual questions, rephrasing, brainstorming, and outline-level drafting are all well within what the lightest model in a current family can do well. Where Luna is thin is on tasks that need careful chain-of-thought or hold up under follow-up scrutiny: legal clauses, multi-step math, code with subtle bugs, anything where a confident-sounding wrong answer is worse than a slow right one.
Where Free still ends
The limits that remain on Free are concentrated in the tools, not the chat:
- File uploads are rate-limited per account per day.
- Image generation has a separate budget that resets weekly and is smaller than Plus.
- Voice mode is gated by a per-account minute budget that runs out faster than most users expect.
- Code interpreter and data analysis tools are unavailable on Free; you cannot upload a spreadsheet and ask the model to analyze it.
- Custom GPTs are not creatable on Free, though they are usable if someone else built them.
A user who sticks to text questions gets the full Free tier. A user who reaches for any of the above tools will hit a wall within the first hour.
Who Free is for after the change
Free with unlimited text is now a reasonable tier for:
- Short drafting tasks (emails, social posts, outline-level writing).
- Quick factual questions where Luna’s depth is enough.
- Brainstorming sessions that produce a lot of low-stakes text.
- Translation and rephrasing tasks that do not need heavy reasoning.
- Educational use cases where the student can tell when an answer is too thin.
Free is still the wrong tier for:
- Document analysis where you need the model to read a long PDF carefully.
- Coding sessions that require running or testing code against a real interpreter.
- Image-heavy workflows, including image generation, image understanding at scale, and any vision-heavy multimodal work.
- Anything where a shallow answer has real consequences (medical, legal, financial, anything that ends in a published document under your name).
The shape of the recommendation has not changed. The ceiling of what Free can do has lifted, but the floor of what Free cannot do is in the same place.
Trade-offs
The change makes Free a stronger default tier for short text tasks, which is the use case OpenAI is most worried about losing to other free chat products. The risk is that the lightest model becomes the de facto answer for users who would benefit from a stronger one. The “Think” button is the bridge, but most users will not find it without a hint.
For paying users, the change does not change the calculus. Plus, Team, and Pro still buy the same things they bought last week: stronger models, more tool calls, longer context windows. The cost of those upgrades is unchanged.
For users coming from another free chat product, the practical question is whether GPT-5.6 Luna on text-only is better than what you are already using. That comparison is personal and worth running on a representative set of your own questions before committing. OpenAI is not the only free chat product on the market, and the rate limit was not the only reason some users left for alternatives. Luna’s quality on text is the real test, and it is one a single week of use will settle.
The timing of the change also matters. OpenAI is removing a throttle right as competitive pressure from other free chat products has grown. The change is good for users, and it is also the right defensive move from OpenAI’s side. Read the announcement in that context.
What I would tell past me
If you stopped using ChatGPT Free because the rate limit felt punitive, give it another week. Luna on text is faster than the Instant model it replaces, and the unlimited cap removes the worst part of the old experience. If you stopped because Free was too thin for the kind of work you do, the change does not fix that. Pay for Plus, or use a different tool.