No typo, no marketing exaggeration: in early August, OpenAI actually killed the message counter for free chats. If you use ChatGPT without a subscription, you can now type as much as you want — at least in plain text conversations.
What's actually changing
Since early August, GPT-5.6 Luna has been the new default model for Free and Go users. Luna is the lean, fast variant of GPT-5.6 — tuned for low latency rather than maximum accuracy. That's exactly what makes it cheap enough to hand out without a cap. OpenAI also says Luna makes noticeably fewer factual mistakes than its predecessor — don't take that as a blank check for trust, though, more on that below.
There's also a new "Think" button: for tricky questions, you can ask for extra reasoning time instead of getting the fast default answer. Handy when you notice ChatGPT is staying shallow — say, on a multi-step math problem or a question that needs actual thinking through, rather than a quick typed reply.
Fewer mistakes doesn't mean no mistakes. Luna is still a language model that recognizes patterns, not a fact database. If an answer feels off or suspiciously smooth, a second source is still worth checking — that was true before Luna, and it's true now.
What still costs money
The generosity has limits, and they matter. Unlimited applies specifically to text chat. Image generation, file uploads, voice chats, and access to Sol — the larger GPT-5.6 model tuned for reliability — remain restricted or subscription-only. OpenAI also notes that abuse safeguards still apply: if you obviously try to hammer the system, you'll eventually hit a wall anyway.
In short: "free" here means "free to talk," not "free for everything." You've probably seen this pattern before — practically every major provider has a free tier, just with clear edges. Inference — the actual computing that happens on every single answer — costs real money, even for a lean model like Luna.
Why now?
ChatGPT reportedly has around a billion weekly users at this point — and the pressure keeps rising anyway. The competition still caps its free tiers: Anthropic doesn't publish a fixed number for Claude at all, just an allowance that refills every five hours, and Google talks about daily limits for Gemini. Both are fine for trying things out and get tight fast in daily use. So OpenAI is going further than anyone else here — and cheap open models out of China like Qwen or Kimi show on top of that that strong AI doesn't need a paywall at all. A free user who hits a limit and just switches to the competition is a lost user for OpenAI. So this change is less a gift than a land grab: whoever scraps the counter first makes it expensive for everyone else not to follow.
What you can try right now
You don't need to do anything new. Just log in at chatgpt.com or open the app — no subscription, no credit card. Luna runs automatically in the background as long as you're not on a paid plan. Ask a few questions in a row without watching the clock, and try the Think button on a tricky one. The difference from the fast default answer is often noticeable.
Who is this actually for?
For beginners and casual users, this is a clear improvement: no more frustration when a conversation abruptly hits a wall. For anyone doing heavy image work or needing maximum reliability — fact-checking for a job, say — a subscription still matters, since Sol and image generation stay out of reach.
The more interesting takeaway: competition for free users has gotten fierce enough that "free, but capped" is starting to look outdated, at least for plain text chat. Whether the same happens for images and video is a different question — and probably still a way off.