OpenAI quietly removed the message-rate caps on ChatGPT Free and Go accounts this week, which is the change most casual users will care about. The headline is “unlimited text chat,” but the fine print is the more useful part of the announcement, because it shifts what “free” means in a way that is easy to miss until you hit it.
What changed
Free and Go users used to be rate-limited to 10 to 40 text messages per 3 to 5 hours on flagship models, with users bumped down to a lighter model after they hit the cap. The new policy drops the message cap entirely for text-only chat. That is the headline.
The new default model for Free and Go is GPT-5.6 Luna, which replaces GPT-5.5 Instant as the lightest model in the GPT 5.6 family. OpenAI’s framing is that Luna is faster, more reliable, and more focused than the model it replaced. From our read, that is what they would say about a default-tier model in any case, so the framing alone is not a strong signal. The real signal is what the model does at the edge case, which is where most of the practical differences show up.
The change was announced to take effect the week of the announcement, so Free and Go users should already see the new default model and the new limits.
What the fine print still restricts
The “unlimited” framing is for text only. PCWorld’s coverage, which is the source this article draws from, calls out the features that still count against quotas:
- File uploads
- Image attachments
- Image generation requests
- Voice mode
A Free user who is summing up a long PDF or generating diagrams will still hit a rate limit, and the limit is not the same shape as the text limit. That is the gotcha.
The change is also specifically about text chat. Other surfaces of ChatGPT, including the voice mode and the image tools, still have separate rate limits they have always had. From our read, that is the most-missed part of the announcement, because the “unlimited” headline reads as broad as the platform.
The “Think” button is the other half of the announcement
OpenAI also added a “Think” button that tells the model to take more time when formulating its response. The intent is deeper reasoning and better answers when the user needs them, rather than a fast answer that is going to be wrong on a hard question.
The button is a small UI change that pairs naturally with the unlimited-text policy. A user who is doing a quick summarize does not want to wait, and a user who is debugging a proof wants to wait. The Think button lets the user pick. From our read of how similar toggles have worked in other chat surfaces, that is a healthier pattern than a “reasoning tier” subscriber lock-in.
A useful way to think about the button: it is the same split that search engines have had for years between fast lookup and deep search. The user signals which one they want, and the system shifts. The chat model equivalent of that is what Think is meant to be, and from our read of the public write-ups, the implementation lives up to that framing. A user who turns Think on for a hard problem and off for a quick task gets the right amount of effort for each.
The cost of the button is latency, not a quota. Users who hit the button on a hard prompt will wait longer, and users who keep it on by default will see every response slow down. From our read, the default state of the button matters more than the announcement makes it sound, because casual users will not change it.
What this means for everyday users
For users who spend most of their ChatGPT time on text prompts, the practical change is that the workflow stops surprising them. The 10-40 messages per 3-5 hours cap was a real constraint that broke the workflow for anyone iterating on a long prompt. The new policy removes that constraint. A user who is drafting an email, debugging a code snippet, or running a back-and-forth on a writing task will not see the floor drop out from under them after a few turns.
Users who use the platform for file analysis, image generation, or voice mode will see a smaller practical change. Those features still have rate limits, and the limits are not necessarily looser than before. The differential between “unlimited text” and “still-limited everything else” is large enough that a user who is paying for ChatGPT for the heavy features will not benefit much from the rate-cap removal. They were not the user the policy was written for.
Users on ChatGPT Plus or the higher tiers can ignore this announcement entirely. The rate caps that just dropped were never on the higher tiers, and the new default model is below the Plus tier anyway. The change is a Free and Go rebalancing, not a broader market reset.
A useful pattern: a user who wants to understand whether the change matters for them should map their actual usage. If 80 percent of the prompts are text and the other 20 percent is file or image, the change is a real win. If the usage is the other way around, the change is mostly cosmetic. From our read, most casual users are in the first bucket, which is why the headline is what it is.
Trade-offs
The change is a clean win for text-only users, but it carries a few caveats worth flagging:
- The default model is lighter than the paid tiers. GPT-5.6 Luna is the lightest model in the GPT 5.6 family, and Free and Go users now default to it. Users who want the smart model have to deliberately route around the default, which most casual users will not do.
- The “Think” button depends on the model. A user who is on Luna and wants deeper reasoning is asking Luna to behave like a heavier model. The button is a UX improvement, not a free upgrade to a bigger model.
- The unlimited-text policy is competitive. Anthropic, Google, and other frontier providers have been offering more permissive free tiers. The structural change OpenAI is making here is to keep Free and Go competitive, not to give away access that they were not under pressure to give away.
- File, image, and voice limits vary in shape. PCWorld’s coverage flags the rate limits on these features, but it does not specify the post-change limits. From our read of the public discussion, the limits are not standardized, and a user who hits them will only know the limit by hitting it.
- The change is fresh. OpenAI announced this with a “starting next week” rollout. The first few weeks of any new policy are when the rate limits, error messages, and model behavior stabilize. From our read, it is reasonable to expect the first month to be a little uneven.
The honest summary
If you are a Free or Go user who lives in text chat, this is the change you have been waiting for. The rate cap is gone, and the workflow stops breaking at the worst moment. If you are a Free or Go user who lives in file uploads, image generation, or voice, the change is mostly noise. The Think button is a nice UX addition that pairs with the new policy, and it is worth turning on when the question is hard enough to wait for. The headline is correct, but the fine print is what determines whether the change matters for you.