Free Version of ChatGPT: What's Included and What's Not

The free version of ChatGPT changed substantially on 6 August 2026, and most guides written before that date are now wrong about the two things people care about most: which model you get, and how many messages you can send.

In short: the default free model moved from GPT-5.5 Instant to GPT-5.6 Luna, and the roughly ten-messages-every-five-hours cap on text chats was removed entirely. The free plan is now considerably more generous than it has been at any point since ChatGPT launched. What has not changed is that free accounts get one model rather than a choice of them, that everything other than text still has limits, and that some capabilities remain behind a subscription.

This guide covers what the free plan actually includes now, which limits survived, what sits behind Plus and Pro, and when paying still makes sense. Rollouts of this kind are staged rather than instant, so if your account still behaves like the old one, it has probably not reached you yet.

Free and Go accounts now default to GPT-5.6 Luna, the smallest model in OpenAI's GPT-5.6 family. It replaced GPT-5.5 Instant, which had been the free default since May 2026. Luna is built for low latency and low cost rather than maximum reasoning power - it is the model designed to answer a great many everyday questions quickly, not the one built for the hardest problems.

OpenAI's own internal testing claims factual errors are around 62 per cent less common in Luna than in the model it replaced. That is the company's figure rather than an independent benchmark, so treat it as a direction of travel rather than a measurement, but the upgrade is real.

Free accounts also get a Think button, which asks Luna to spend longer on a harder question. That is the free tier's single reasoning setting - paying accounts get a slider that controls how much reasoning effort each answer receives. The larger models in the family, including GPT-5.6 Sol, are not available on the free plan at all.

Where things stand as of August 2026. OpenAI states plainly that the default model and available limits change over time, so verify anything you are planning around against their current documentation.

For ordinary writing, questions, explanation and research, the free plan now does most of what a casual user needs. Unlimited text was the constraint that pushed people towards a subscription, and removing it changes the calculation for a large number of users who were paying purely to avoid hitting a wall mid-task.

• Unlimited text conversations. The headline change. You can keep a long conversation going without rationing messages or waiting for a reset window.
• Web search. Free accounts can look things up rather than relying only on training data, which matters for anything recent.
• Image generation and file uploads. Both available, both with their own separate caps that the unlimited text change did not touch. Uploaded files go to a library you can reuse, with a storage limit on free accounts.
• Voice mode. Available, defaulting to a smaller voice model than paying accounts receive.
• Existing custom GPTs. You can use ones other people have built. Creating and publishing new ones is not available on personal accounts of any tier, including paid ones.

The word unlimited applies to text and nothing else. If your work is mostly generating images, analysing documents or using voice, the change affects you very little.

• Every model except Luna. This is the most significant restriction and the one least discussed. Free accounts get one model. The larger members of the family, including Sol, require paying, and there is no way to pick a different model for a task that would suit one.
• The reasoning slider. Paid accounts control how much reasoning effort goes into each answer. Free accounts get a single Think button and no control beyond on or off.
• Deep Research. The long-running research mode that works through many sources and produces a cited report stays on paid plans.
• Agent Mode and Codex. The autonomous task and coding environments are not part of the free plan.
• An ad-free experience. Free users in some countries, including the United States, see advertising in the interface. Paying removes it. This is a real difference in daily use that rarely appears in feature comparisons.

With text limits gone, the question has changed. It is no longer how much can I use it but which model do I get and what can it do. Check current prices on OpenAI's own page, since they vary by country and change.

Plus is worth paying for if you regularly need a stronger model, work with a lot of files or images, or want the ads gone. If you mostly type questions and read answers, the free plan now covers it - which is a sentence that was not true three months ago.

Free ChatGPT gives you unlimited access to one model. That is a genuine upgrade, and it is also the shape of the remaining limitation: you cannot choose. Luna is built for speed and cost rather than for hard problems, and when a task would suit a reasoning-tier model, a long-context model or an image model, the free plan has no answer other than upgrade.

That matters more than it sounds, because models differ in ways that show up on ordinary work. A reasoning model shows its steps on a multi-stage problem. A long-context model holds a whole document rather than the first third of it. Models from different labs are trained and tuned differently and produce noticeably different answers to the same question - which is why running one prompt through two of them is the fastest way to find out which suits your work.

This is the gap Chat Smith is built for. Rather than one vendor's model, it puts several in one place - GPT tiers alongside Claude, Gemini, Grok and DeepSeek - so you can match the model to the task instead of accepting whichever one your plan includes. The full model list shows what is available, and there is a free tier to test the idea before deciding anything.

The cost comparison is where this gets concrete. ChatGPT Plus is $20 a month. Chat Smith Pro is $9.99 a month, or $39.99 for a year - roughly $3.33 a month if you commit to twelve. For that you get GPT-5.6 Sol alongside Claude Sonnet 5, Gemini 3.5 Flash, Grok 4.5 and DeepSeek V4, rather than one vendor's family. Prices were correct at the time of writing and are on the Chat Smith pricing page.

The free tier is genuinely usable rather than a trial, and two things on it are worth noting because ChatGPT charges for them: Deep Research with live web search, and image generation on ChatGPT Images 2.0 and Nano Banana Pro.

To be clear about where it does not help: if you only ever type text questions and one model's answers suit you, free ChatGPT is now an excellent deal and there is no reason to add anything. The case for a multi-model tool is a case about variety of work, not about message caps - those are gone.

For a large number of people the honest answer is now no. If you use ChatGPT for writing, questions, explanation and looking things up, the free plan covers it and the main reason people used to upgrade has been removed. Try the free tier properly for a fortnight before assuming otherwise.

Pay if one of these applies: you regularly hit the image or file caps, you need Deep Research or the coding and agent environments, the ads bother you, or your work is hard enough that the smallest model in the family is visibly struggling. That last one is the real test, and you will know within a week of trying.

And if what you want is access to more than one model rather than more of one, that is a different purchase - and at $9.99 a month against ChatGPT Plus at $20, the arithmetic is worth doing before you renew anything. Test it on your own work first: the free tier is enough to find out whether model choice actually changes your results, which is the only question that matters. Our guide to the best AI chatbot compares the alternatives, and GPT-5.6 Sol covers what the paid OpenAI tier actually gives you.