Analysis
OpenAI and Anthropic are both cutting prices to hold onto cost-sensitive customers who are increasingly testing cheaper Chinese alternatives, according to Ars Technica's reporting. OpenAI cut the price of GPT-5.6 Luna -- its fastest and most affordable model -- by 80%, and Anthropic introduced a new Claude Opus 5 tier priced at $5 per million input tokens and $25 per million output tokens, a meaningfully lower entry point than its prior flagship pricing.
## The scale of the Chinese undercut The pressure forcing these cuts is specific and quantifiable: DeepSeek, Zhipu AI's GLM-5.2 and Moonshot's Kimi K3 are priced 60-90% below comparable Anthropic and OpenAI models on a per-token basis. That's not a marginal discount aimed at price-sensitive hobbyists -- it's a gap large enough to change enterprise vendor-selection math for any company running high-volume inference workloads, where per-token cost compounds quickly at production scale. Alibaba's Qwen family topping 3 billion downloads this same week is the distribution-side expression of the identical dynamic: developers choosing free or dramatically cheaper models for workloads where frontier-grade quality isn't strictly necessary.
“## The quality counterargument The price war isn't a clean story of US labs simply losing ground, though.”
## The quality counterargument The price war isn't a clean story of US labs simply losing ground, though. A study dated August 13 found that Anthropic's models can still deliver a lower total cost of ownership than leading Chinese open-weight alternatives for specific tasks, despite charging more per token -- because Anthropic's models complete tasks in fewer total tokens, producing higher-quality output on the first attempt rather than requiring multiple retries or longer reasoning chains. That distinction matters for how enterprises should actually be comparing vendors: sticker price per token is the easiest number to compare, but it's not the number that determines actual workload cost once retry rates and output quality are factored in.
## What the price war means for both labs' growth story The counterweight to treating this purely as a defensive retreat: both OpenAI and Anthropic are reporting record enterprise revenue in the same week they're cutting prices. OpenAI's CFO disclosed enterprise revenue has overtaken consumer revenue at a roughly $40 billion annualized pace, and Anthropic reported preliminary Q2 revenue above $11.5 billion. Cutting per-token price while growing total revenue means usage volume is expanding fast enough to offset the lower unit economics -- a sign both labs believe volume growth from broader accessibility outweighs the margin given up on each individual token, at least for now.
The open question is how long that trade stays favorable. If Chinese models continue closing the quality gap that currently justifies Anthropic's and OpenAI's premium pricing, the total-cost-of-ownership argument that's currently protecting US labs' margins gets weaker every quarter, and price cuts that look like a controlled, strategic response today could become a much more defensive scramble within a year.
