OpenAI and Anthropic – two of America’s artificial intelligence (AI) pioneers – have slashed access fees for their language models to stop budget-focused enterprise clients from migrating overseas even as they face mounting pressure from low-cost Chinese rivals. The price battle arrives as mounting computing expenses force corporations across Europe and Silicon Valley to curb their usage. This has allowed Chinese developers like DeepSeek and Moonshot to win over major business accounts.Citing Silicon Data’s token price index, the Financial Times reported that rates paid by consumers for flagship US models have plummeted nearly 25% since mid-July. The downward shift highlights a pivot for domestic developers, who long relied on proprietary, closed architectures and competed primarily on raw technical power rather than pricing.
Cuts for mid-tier AI models
The recent US developer price cuts are mainly on intermediate products, which are designed to be as economical as open-weight alternatives. OpenAI has cut the price of access to its fastest and cheapest option, GPT-5.6 Luna, by 80%. Input token rates plummeted from $1 down to $0.20 per million, while output charges dropped from $6 to $1.20 per million.Similarly, Anthropic launched Claude Opus 5 at $5 per million input tokens and $25 per million output tokens, offering top-tier reasoning at exactly half the rate of its premium Fable 5 tier. Additionally, the startup abandoned a price hike on its Sonnet 5 model that was scheduled for September.
Corporate budgets drive adoption of Chinese silicon
The pricing battle unfolds as both Anthropic and OpenAI prepare for initial public offerings (IPOs) at trillion-dollar valuations, all while institutional backers demand clear paths to profit on massive infrastructure expenditures.Simultaneously, both US providers have moved business clients away from flat-rate subscription agreements toward usage-based billing, charging enterprises directly for the computing power they utilise. In response to rising overhead, commercial customers such as Airbnb and DoorDash have started deploying Chinese-built models to contain operational expenses.These decisions coincide with rapid advancements from Chinese labs, which have sharply narrowed the technical performance gap against top-tier American systems while allowing developers to freely download, run, and modify their model weights.