OpenAI and Anthropic slash mid-tier model prices as Chinese rivals win

👁 115 reads

OpenAI has cut the price of its GPT-5.6 Luna model by 80 percent, while Anthropic has launched Claude Opus 5 at half the price of its most capable model, as both companies respond to users switching to cheaper Chinese AI alternatives.

The cuts are a direct reaction to cost pressure on corporate customers. As Ars Technica reports, token prices from leading US labs have fallen by almost a quarter since mid-July, according to Silicon Data’s token price index. Chinese developers including Moonshot and DeepSeek have been gaining ground with users across Silicon Valley and Europe, helped by capable open-source models that can be freely downloaded and customised.

OpenAI reduced GPT-5.6 Luna - described by the company as its “fastest and most affordable model” - from $1 to $0.20 per million input tokens and from $6 to $1.20 per million output tokens. Anthropic priced Claude Opus 5 at $5 per million input tokens and $25 per million output tokens, describing it as offering “frontier intelligence… at half the price” of Fable 5. Anthropic also this week cancelled a planned price increase for its Sonnet 5 model that had been scheduled to take effect from September.

The pricing pressure is partly structural. Both OpenAI and Anthropic have been shifting enterprise customers away from flat subscriptions toward usage-based billing, a move that has pushed some businesses to cap AI usage or test alternatives. Companies including DoorDash and Airbnb have publicly said they have begun using Chinese-made models to control costs.

The competitive backdrop matters for how US labs are positioning themselves. Both OpenAI and Anthropic are planning initial public offerings at trillion-dollar valuations, and investors want evidence that heavy spending on AI infrastructure can generate returns.

Practitioner takeaway: Developers and businesses currently on usage-based billing should re-examine which model tier they are actually routing tasks to. The latest cuts apply specifically to mid-tier products - not the flagship models - so high-volume, lower-complexity workloads such as classification, summarisation or first-pass drafting could now run significantly cheaper on US-lab infrastructure than they did a month ago. Headline token prices still require care: a more capable model completing a task in fewer tokens or fewer attempts can cost less in practice than a cheaper model that needs more tries.

// read moreOpenAI closes gap with Anthropic among US business spenders, Ramp dataMagazineSpotify to label AI artists, drop from recommendations mid-SeptemberAI ToolsOpenAI Disbands Preparedness Team, Splits Work Across GroupsNews

Kinetyra