GPT-5.6 Luna has become the default model for ChatGPT's free tier, with unlimited exchanges, following a price cut of around 80% on the API. A recently developed model accessible without counting and without paying represents a major shift in accessibility. It deserves to be welcomed, and its funding deserves to be understood.
We explained where this price cut came from: runtime optimisations to which the model itself contributed. Here is its most visible consequence for the general public.
What unlimited free access really changes
The main change is psychological as much as practical. As long as a quota exists, you ration your questions. You hesitate to ask for a rewrite, you don't follow up to go deeper, you save your messages for what matters.
That mental economy disappears. And it changes the nature of use: you move from a tool you consult to a tool you work with, iterating. And it is precisely in iteration that these systems become useful, as we explained in our article on crafting requests.
The accessibility effect is also real. For a student, someone without a budget, a country where a monthly subscription in a strong currency is out of reach, the difference between limited free and unlimited free is considerable. This is an extension of what we described regarding accessibility.
How it is funded
The question is not cynical; it is necessary to understand what might change next. Three mechanisms combine.
The real drop in costs. Inference optimisations have divided the unit cost by five. What was untenable becomes sustainable, as we explained in our article on the two costs of AI.
Conversion. A generous free tier feeds a paid one. The more people use it, the more some hit a limit that justifies a subscription.
Strategic positioning. In a market where capabilities are converging and where a competitor has crossed a billion users thanks to its distribution, holding the default position is worth a lot. We wrote about this regarding the battle for the interface: whoever owns the relationship with the user decides what runs behind it.
A free tier usually comes with different terms on data. We detailed this in our article on what your AI knows about you: on consumer plans, conversations may be used to improve models unless you opt out, whereas professional plans commit contractually otherwise. That is not a reason to avoid it; it is a reason to know what you put into it. Timing matters too: this free access arrives as an IPO is being prepared, where user numbers are an asset presented to investors.
The question of dependency
There is one point we have learned to watch. Sustained free access builds habits, both among individuals and in organisations that build on top of it.
And we saw this summer that prices go back up: a 50% promo ending, a 93% increase. Nothing guarantees that an unlimited free tier stays that way, and the history of online services suggests the opposite.
That does not condemn using it. It simply invites you not to build something critical on a free tier, and to know what you would do if it ended.
What to take away
This is good news, and it would be churlish to present it otherwise. Hundreds of millions of people now have unlimited access to a tool that required a subscription six months ago. The access gap that was widening between those who paid and everyone else is partly closing.
What remains to watch is not the sincerity of the move, but its durability. A free tier funded by a real drop in costs can last. A free tier funded by a position to be won lasts until the position is secured. We will know which one it was by looking, in two years, at what is still free.