ChatGPT's Free Tier Just Went Unlimited — Here's My Honest Take on What Actually Changed

Two days ago OpenAI quietly stopped rationing the free version of ChatGPT. No keynote, no launch event — just a blog post and a rollout that started landing in accounts mid-week. I've been living in the free tier since it reached mine, deliberately not touching my paid account, to find out if "unlimited" actually means what it says.

Here's the announcement in one sentence: GPT-5.6 Luna is now the default model for Free and Go users, replacing the older GPT-5.5 Instant, the daily message caps on ordinary text chat are being lifted, and free users get a new Think button that gives the model extra time to reason through a harder question before answering. Plus and Pro subscribers get their own upgrade — a retuned GPT-5.6 Sol with a reasoning slider — but that's not the part that surprised me. The free-tier change is the one worth actually testing, because it's the tier almost everyone reading this has an account on.

What "unlimited" turned out to mean in practice

I spent two full days doing exactly what I used to ration on the free plan: long back-and-forth conversations, follow-up questions I'd normally save for a fresh chat to avoid hitting a cap, and the kind of "actually, wait, what about—" thread that used to burn through a daily allowance by lunchtime. I never once saw the old "you've reached your limit, try again later" message. That message has genuinely been the single biggest reason I'd tell people ChatGPT's free tier wasn't worth building a habit around — it made the product feel like a trial rather than a tool. That specific friction is gone.

What hasn't changed: image generation, file uploads and some of the more agent-style tools still carry separate limits, free or not. This is a change to text chat specifically, not a blanket "everything is free now" announcement, and it's worth being precise about that before anyone cancels a subscription on the strength of a headline.

The model swap matters more than the limit

The quieter change is arguably the bigger one: the free tier isn't just uncapped now, it's also running a different, newer model. GPT-5.6 Luna replaces GPT-5.5 Instant as the default, and OpenAI's own internal testing claims a 62% drop in factual errors compared with the model it's replacing. I want to be upfront that this is OpenAI grading its own homework, not an independent benchmark, so I tested it the only way that actually means anything to me: I asked it the same handful of specific, checkable questions I keep in a running list for exactly this purpose — release dates, technical specs, "who currently holds X record" style questions where I already know the right answer. Luna got noticeably more of them right than I remember the old free default managing, and where it wasn't sure, it said so instead of inventing a confident-sounding number. That second part is the one that actually changes how I'd use it day to day.

The Think button, and when I actually reached for it

Free users now get a Think button that gives the model extra time to reason before it answers, which used to be something you only got with a paid reasoning model. I tried it on the kind of question I'd normally have felt was overkill for a free tool — working through a multi-step comparison with real trade-offs rather than a single factual lookup — and the difference was obvious: slower, visibly working through the problem, and landing on an answer that held up better under a follow-up push. I wouldn't reach for it on a quick question; it's genuinely slower and mostly unnecessary there. But for the harder end of what I'd ask a beginner-friendly AI tool to do, it closes a gap that used to require paying.

"The uncapped chats are the headline. The model swap and the Think button are the part that actually changes what the free tier is good for."

So — is Plus still worth paying for?

Honestly, for the way I use ChatGPT day to day, this update just made that a much harder yes. I compared how a paid GPT-5.6 Sol conversation and a free Luna conversation handled the same three real tasks from my week — drafting an email, summarising a long document, and untangling a scheduling conflict — and the free version was close enough on all three that I wouldn't have noticed the difference blind. Where Plus still earns its keep is the reasoning slider on Sol, higher caps on image generation and file uploads, and priority access if the free model ever queues under load, which I haven't hit yet but expect to during a busy US morning. If your usage looks like mine — mostly conversational, occasional document work, rarely generating images — I'd genuinely try a week on the free tier before renewing anything. If you lean on file uploads, images or the deeper reasoning tools constantly, the case for paying hasn't gone away. Either tier, the bigger daily win for recurring work is still setting up Projects properly rather than starting every chat from zero.

Why this is worth reacting to at all

Free-tier changes rarely make me want to write anything — most of them are minor rebalancing that nobody notices. This one is different because it removes the exact friction that made the free tier feel disposable rather than dependable, and pairs it with a genuinely newer model rather than just more of the old one. I've written before about how ChatGPT and Claude actually differ in daily use, and this update shifts that comparison meaningfully for anyone weighing up a free account against a different tool entirely — the free version just got harder to dismiss. If you're building out an AI toolkit and want the fuller landscape rather than just this one update, the AI Tools hub has the honest, tested rundown of what I actually reach for each week.

Free-tier changes like this one are the reason a price list goes stale so fast. I keep a running page of what the major AI tools actually cost, rechecked monthly, so the comparison is against today's pricing rather than last quarter's.

What I'm doing next

I'm staying on the free tier for another week before deciding anything about my own subscription, mostly because two days isn't long enough to know whether the "unlimited" claim holds up once OpenAI's servers are under real peak load rather than a quiet Thursday. I'll update this if that changes. In the meantime, if you want the next AI update I actually test rather than just repeat from a press release, join the Brite newsletter — no hype, just what held up.

Questions people actually ask

What actually changed in the ChatGPT free tier update?

OpenAI made GPT-5.6 Luna the default model for Free and Go users, in place of the older GPT-5.5 Instant, and is removing the daily message caps that used to cut a long free conversation short. Free users are also getting a Think button that gives Luna extra time to reason through a harder question. Image generation, file uploads and some advanced tools still carry separate limits — the change is specifically to text chat.

Is ChatGPT completely free now?

No — it's the rate limit on ordinary text conversations that's gone, not the whole product. You can still hit walls on things like image generation, file uploads, and the more advanced agent-style tools, which remain capped or paid-only. Think of it as: the everyday chat use case just stopped being rationed.

Is ChatGPT Plus still worth paying for after this update?

For pure back-and-forth chat, the gap just got a lot smaller. Where Plus still earns its keep is GPT-5.6 Sol's reasoning slider, higher limits on file uploads and image generation, and priority access when the free model is under heavy load. If you mostly ask ChatGPT questions and rarely touch its other tools, I'd genuinely try living on the free tier for a week before renewing anything.

Does GPT-5.6 Luna hallucinate less than the old free model?

OpenAI's own internal testing reported a 62% drop in factual errors for GPT-5.6 Luna compared with the GPT-5.5 Instant model it replaces as the free default. That's OpenAI grading its own homework, so I wouldn't repeat it as settled fact, but it matches roughly what I noticed in two days of testing: fewer confidently wrong answers on specific, checkable questions.