OpenAI just cut its cheapest model's price by 80% overnight
On 30th Jul’26, three weeks after launching the GPT-5.6 family, OpenAI cut the price of Luna, its fastest, cheapest tier by 80%, to $0.20 per million input tokens and $1.20 per million output. The mid-range Terra dropped 20%, to $2/$12. The flagship, Sol remained unchanged. OpenAI says it's passing on efficiency gains from running the models more cheaply.
What it actually means
The interesting part isn't the discount, it's who's discounting. For a month the cheap-model story has been a Chinese one; now the US incumbent has reacted and cut its budget tier, weeks after launch. That isn't generosity, it's pressure the price of routine AI work is being competed to the floor, and the labs are absorbing it. For anyone building, the practical read is the same as it's been all month, only cheaper: the boring 80% of a workflow, classifying, drafting, running the tests will now costs almost nothing, whoever you buy it from. Two moves this week: re-price your heaviest high-volume job against the new Luna rate before you assume your current bill is fixed, and keep the expensive flagship for the hard twenty per cent where it actually earns its keep. The models you already chose didn't change. The maths underneath them did.
The cheap-AI story stopped being only Chinese, the US incumbent just reacted and cut its budget tier 80% to keep up.
80%OpenAI's overnight price cut on its cheapest model
Source: OpenAI · Friday 31 July 2026