OpenAI and Anthropic are cutting prices on their flagship artificial intelligence models as cheaper Chinese rivals win over cost-conscious customers, turning what had been an arms race in model quality into a fight over bills. OpenAI has cut GPT-5.6 Luna pricing by 80%, while Anthropic has launched Claude Opus 5 at half the price of its previous flagship. The move comes as both US labs still prepare IPOs at trillion-dollar valuations, and as investors watch for proof that huge AI spending can translate into durable returns.
The shift is happening fast enough to show up in the market itself. Silicon Data’s token price index shows prices customers pay for leading US lab models have fallen almost a quarter since mid-July. That drop suggests pricing pressure is not just a marketing tactic but a real change in how the biggest AI companies are selling access. It also points to a broader adjustment in the industry: premium model makers are learning that expensive does not always mean sticky, especially when buyers can compare costs across more providers than ever.
OpenAI’s latest move is the clearest sign of the new mood. The company cut GPT-5.6 Luna pricing to $0.20 per million input tokens from $1, and to $1.20 per million output tokens from $6. That is an 80% reduction on the input side and a sharp reset for customers who have been trying to rein in usage bills. Anthropic, meanwhile, introduced Claude Opus 5 at $5 per million input tokens and $25 per million output tokens, which is half the price of its Fable 5 flagship. It also cancelled a planned Sonnet 5 price increase that had been due in September.
The pricing changes matter because they show both companies defending the high end of the market while also trying to avoid losing price-sensitive enterprise users. Mantas Lukauskas, AI tech lead at Hostinger, put it bluntly: “The US labs have cut the middle and are defending the top.” The line captures the strategy emerging across the sector. The best-known models still command premium prices, but the companies are narrowing the gap in the middle tier, where customers may be most likely to switch if a cheaper rival looks good enough.
The immediate pressure is coming from Chinese models, especially Moonshot and DeepSeek. The Financial Times reported that OpenAI and Anthropic are releasing cheaper models to retain customers switching to those rivals. DoorDash and Airbnb said they have started using Chinese-made models to rein in bills, a signal that the appeal of lower-priced alternatives has moved beyond theory and into corporate procurement. For major software users, the decision is no longer just about benchmark scores. It is about how much model access costs when scaled across large teams and high-volume workflows.
That does not mean the US labs have lost their edge. But it does mean the competitive backdrop is changing. Open-source and lower-cost alternatives are improving fast enough to force buyers to compare models more aggressively than before. Wang Tiezhen, an independent AI consultant and former head of APAC ecosystem at HuggingFace, said: “Open-source is closing the performance gap with closed models faster than anyone expected.” That trend gives enterprise buyers more leverage and makes it harder for any one lab to hold pricing power for long.
The money question is becoming more urgent because the economics of AI usage are shifting as well. OpenAI and Anthropic are moving some enterprise customers away from flat subscriptions and toward usage-based billing. That switch can help companies better match revenue to compute costs, but it also exposes them to demand swings and can reduce the predictability that investors usually prefer. It is another sign that the business model is still being built in real time, even as the biggest players talk like future public giants.
The timing is awkward for both firms. They are preparing IPOs at trillion-dollar valuations, but the market is now asking whether the scale of investment in frontier AI will produce returns fast enough to justify those numbers. Cheaper model pricing can help retain users, but it can also make it harder to widen margins. In other words, the more competitive the market becomes, the more difficult it may be to turn leadership in AI into the kind of profits public investors typically demand.
Not everyone wants to call it a war. Jack Gold, founder of J.Gold Associates, said: “I hesitate to say it’s a price war because we’re not quite there yet. But it’s certainly a price competition to try and get more users on board.” That distinction matters. A full price war implies a race to the bottom. What is visible now looks more like selective discounting, with the biggest labs trimming costs where customers are most likely to shop around and holding firmer pricing where they still think the product is indispensable.
Even so, the pattern has the feel of an industry under stress. The leading US labs are cutting prices at the same time as competitors in China are gaining ground, customers are more willing to switch providers, and model prices are falling across the board. The market is also getting more granular, with buyers calculating model economics by task rather than by brand. That shift makes pricing far more transparent and far less forgiving for companies that once could rely on prestige alone.
For OpenAI and Anthropic, the risk is that lower prices solve one problem while creating another. Cheaper access can protect market share, but it can also accelerate expectations that model intelligence should become cheaper every quarter. If customers get used to rapid discounts, any attempt to raise prices later becomes harder. Anthropic’s cancelled Sonnet 5 increase is a reminder that even planned hikes can be pulled back when the market turns against them.
The bigger issue is that both companies are trying to prove two things at once: that they can win users in a more competitive market, and that they can still grow into the enormous valuations investors are expecting. Those goals are not always aligned. Cutting prices may slow customer churn, but it also raises the bar for revenue growth. The next test will be whether usage-based billing and premium tiers can offset the lower rates now being used to hold the line.
For now, the signal is clear: the AI model market is moving from swagger to arithmetic. Customers are comparing token costs, enterprise buyers are testing alternatives, and the biggest US labs are responding with sharper pricing. If Anthropic reinstates its planned September price increase, or if OpenAI follows with another round of cuts, the current standoff could harden into something closer to a true price war. For investors, the question is no longer who can build the smartest model. It is who can sell it without giving away the economics.
