OpenAI and Anthropic Cut Prices to Counter Chinese AI, Price War Slashes User Costs by Nearly a Quarter
Nashnova编辑部
OpenAI and Anthropic have slashed mid-tier model pricing in direct response to low-cost Chinese AI rivals — since mid-July, what users pay for leading U.S. AI models has dropped by nearly a quarter, setting up a fight over who controls AI pricing power.
How deep are the cuts?
OpenAI dropped GPT-5.6 Luna input pricing from $1 to $0.20 per million tokens, and output from $6 to $1.20 — an 80% reduction.
Anthropic launched Claude Opus 5 at $5 input / $25 output per million tokens — half the price of its flagship Fable 5. It also scrapped a planned September price hike for Sonnet 5.
This means → both labs are releasing steep mid-tier discounts at the same time, and the target is clear: Chinese open-source models.
Why now? What are Chinese models actually taking?
Models from Moonshot and DeepSeek are spreading from Silicon Valley into Europe. DoorDash and Airbnb have both said they are using Chinese models to cut costs.
Chinese open-source models can be downloaded free and fine-tuned in-house — a powerful draw for cost-conscious companies. Meanwhile, Anthropic and OpenAI have been shifting enterprise clients from flat subscriptions to usage-based billing, driving AI bills higher. Some firms have begun capping usage or testing alternatives.
In plain terms = Chinese models are not just cheaper — they are "download free, customize yourself." That undercuts U.S. mid-tier products first.
Can cheap Chinese models match the performance?
Data from AI benchmarking firm Artificial Analysis shows: Anthropic Opus 5 at "medium" compute roughly matches Moonshot's Kimi K3 at "maximum" compute in both performance and per-task cost.
OpenAI GPT-5.6 Luna at "maximum" compute matches DeepSeek V4 Flash in performance, but costs roughly twice as much per task.
This means → on mid-tier price-performance, Chinese models have caught up or pulled ahead. U.S. labs needed these price cuts just to get back on the same playing field.
Are flagship models getting cheaper too?
No. The cuts are confined to mid-tier lines. Flagship model pricing has held steady or even risen.
Mantas Lukauskas, AI technology lead at web host Hostinger, called the move "the first real test" of whether U.S. labs can hold premium pricing: "They're cutting in the middle and defending the top."
In plain terms = give ground in the middle to keep customers, hold the line at the top to protect margins — whether that defense holds is still an open question.
What other pressure sits behind this price war?
Both OpenAI and Anthropic are preparing IPOs at trillion-dollar valuations. Investors want evidence that massive AI spending can convert into returns.
A person close to Anthropic said the Opus 5 pricing below Fable 5 is "a natural expression of the model-family architecture, unrelated to competitors." Both companies declined to comment.
This reflects a dilemma: mid-tier prices must fall to retain users, but cuts squeeze margins — and margins are exactly what investors need to see. Whether flagship pricing can hold will be the key variable testing that logic.
Content is for reference only, not financial advice.