The AI Price War Just Reached Your Business

Pick almost any market and the story is the same: prices fall when somebody credible shows up to take your customers. Rents, airline seats, cell plans, streaming. Now add AI models to the list.

OpenAI cut the price of two models in its GPT-5.6 family today — the cheapest tier, Luna, by 80%, and the mid-tier, Terra, by 20%. I went and looked at the company's published rate card and API changelog this morning: Luna now runs $0.20 per million input tokens and $1.20 per million output tokens, down from $1 and $6. Terra drops from $2.50/$15 to $2/$12. The changes are live. The top-end model, Sol, didn't move — still $5 in, $30 out.

Here's the argument, and then I'll earn it: the reason a small business in Pawtucket or a software shop on the East Side is going to pay less to run an AI feature next month has almost nothing to do with anyone's feelings about artificial intelligence, and everything to do with three enormous companies clawing at each other's enterprise contracts.

Start with the timing. GPT-5.6 reached general availability on July 9. Three weeks later, two of its three tiers got repriced. That is not a product roadmap. That's a response.

Now walk the chain backward. In February, Anthropic cut Opus pricing by 67%, from $15/$75 per million tokens down to $5/$25. In May, at its I/O conference, Google restructured its consumer subscriptions — the top Gemini tier fell from $249.99 a month to a new $99.99 entry point — and shipped a cheaper Flash model on the API side. On June 30, Anthropic launched Claude Sonnet 5 with introductory pricing of $2 input and $10 output, and its own documentation is explicit that the deal expires: standard pricing of $3/$15 takes effect after August 31. Then OpenAI, in July, cut.

Six price moves in six months. Every one of them downward.

OpenAI's own explanation is about engineering, not rivals. In its announcement the company said its "strategy remains focused on advancing both capability and efficiency so each generation of intelligence can accomplish more work at a lower cost," and pointed to internal gains — roughly a 20% reduction in the end-to-end cost of serving the model and better than 15% improvement in token-generation efficiency. Both things can be true. Efficiency gains give you room to cut; competitors give you the reason.

And the customers gave them the pressure. Enterprise buyers have gotten stingy, demanding clearer returns before deploying expensive models. Uber, by one widely reported account, burned through its entire annual AI budget in four months before putting spending tiers on some tools. That is the demand side of this market talking, and it is saying the same thing your accountant would.

So what does this actually mean here?

Rhode Island is a beneficiary in this fight, not a participant. Nobody in this story is headquartered closer than Mountain View. But the state has a real cluster of companies that buy these tokens by the millions — Utilidata, working on AI for the electric grid; Simantex; HavocAI in maritime autonomy; consulting and development shops like Company 50 and MojoTech that build client systems on top of somebody else's model. When the cheap tier drops fivefold, the arithmetic on which jobs are worth automating changes for all of them.

One caution, and it matters. This is not "AI got cheap." The cuts landed only on the commodity tiers — the models that do classification, extraction, routing, first-pass drafting, and the long agent loops where one instruction triggers dozens of calls. Frontier reasoning didn't budge. And per-token savings evaporate fast if your usage triples, which is exactly what tends to happen when something gets cheaper.

There's a policy footnote worth registering. Rhode Island's statewide AI roadmap, adopted in February, calls for helping small businesses adopt these tools partly through subsidized access. The market just delivered a chunk of that for free — for now, and only because three companies are afraid of each other.

Which brings us back to the calendar. Anthropic's introductory Sonnet 5 rate ends August 31, when it climbs to $3/$15. Watch what OpenAI and Google do in the days after that. Prices fall when somebody credible shows up to take your customers. They stop falling when nobody does.