Corporate America is increasingly choosing cheaper artificial intelligence models rather than automatically paying for the most advanced frontier systems, according to Financial Times reporting. The shift is putting pressure on pricing across the AI industry, as new models close the capability gap while token costs keep falling.
Cost-driven buying behavior. Ramp’s September 9 AI Index, based on August 2026 spending data, found 43.8% of sampled US businesses purchased Anthropic products, compared with 39.8% for OpenAI. The same index noted that rising use of cheaper models is a threat to AI companies that depend on businesses steadily spending more on frontier systems.
Open-weight momentum. AlphaSense data cited by the Financial Times shows executive mentions of open models in corporate events jumped sixfold in August and September compared with the same period last year. AT&T said it runs about 40% of its AI workloads on open models and plans to reach 70% within a year. PNC Financial CFO Robert Reilly said the bank runs frontier models through major cloud providers while also keeping open-weight models in its own data centers. Match Group CEO Spencer Rascoff said the company uses Chinese open-weight models to help manage costs, and Tinder technology chief Vinay Kuruvila said open models may eventually replace frontier models if they keep improving.
New model pricing. OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, with prices at least 50% lower than previous GPT-5.6 models. GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens. Anthropic launched Claude Opus 5.5, which it says costs about 40% less to run than its predecessor by being more token-efficient. In the same week, xAI presented Grok 4.7.
Performance versus price. Zapier’s AutomationBench shows Claude Opus 5.5 leads at 42.47% efficiency for $1.44 per task, while GPT-6 Sol at XHigh achieves 33.2% efficiency for $0.27 per task. Artificial Analysis data shows GPT-6 Sol at maximum effort scores 48 on its Intelligence Index at an estimated $1.06 per task, while Xiaomi’s open-weight MiMo-V2.6-Pro scores 46 at only $0.13 per task. Epoch AI data indicates the price of achieving a constant AI performance level has dropped by roughly 47% quarterly since 2023, or about 13 times yearly.
Lower prices may not mean lower bills. Analysts point to the Jevons paradox, where cheaper resources drive greater consumption. Gartner has forecast a more than five-fold rise in inference costs for agentic workflows by 2028. Digital Realty built an internal chat tool on open models partly because proprietary data never goes into a frontier model, and Anthropic product management lead Dianne Penn said the company is focused on making its models answer using fewer tokens depending on the chosen effort setting.
An OECD agentic AI report published in September 2026, based on interviews with 25 organizations, found cost savings were a common reason for using agentic systems, alongside higher productivity and addressing labor shortages.
