Eighty percent off, and a billion users

A tier cut by four fifths three weeks after launch is not a pricing strategy. It is a response, and the shape of the cut says where the pressure came from.

GPT-5.6 Luna fell from $1/$6 to $0.20/$1.20 per million tokens, Terra took 20%, and flagship Sol held at $5/$30. ChatGPT is reported at roughly a billion weekly users.

Read which tiers moved

Deep discount at the cheap end, nothing at the top. That is not a company repricing its whole line — it is a company defending the commodity end while asserting that frontier capability still commands a premium.

Three weeks after launch is the other tell. Planned pricing curves do not move that fast. Something arrived in the market between launch and the cut, and it arrived below rather than above.

The volume cuts both ways

A billion weekly users is what makes an 80% discount survivable and also what makes it necessary. At that scale a fifth of the previous price is still an enormous business, and being undercut at that scale is existential rather than annoying.

What the labs say the constraint is

Anthropic's co-founder named compute access, not research, as what keeps Claude at the frontier. Three years ago the stated bottleneck was talent. Two years ago, data. Now it is megawatts.

Unlike the previous two, this one cannot be accelerated by hiring. It has a delivery schedule, a permit trail and a grid operator attached — and the first Helios gigawatt is not due until the first half of 2027, several model generations away.

Which is the honest connection between a price cut and a substation. If compute is the binding constraint and compute takes years, then price competition is what happens in the meantime, among companies that have already bought everything available.

CNBC — OpenAI cuts prices for two of its GPT-5.6 AI models as companies grow sensitive to costs → · VentureBeat — AI price wars: OpenAI cuts GPT-5.6 Luna prices by 80% → · Pulse 2 — AMD and Anthropic partner on 2-gigawatt AI infrastructure deployment →