OpenAI and Anthropic are pushing the industry to shift the metric for AI model costs from the traditional "cost-per-token" to "cost-per-task." This move aims to counter market competition from companies like Meta and SpaceX, which offer extremely low token prices, and to re-establish a value framework for their own high-premium models. Both OpenAI CFO Sarah Friar and Jonathan Pelosi, Head of Financial Services at Anthropic, stated that lower-cost models might require more attempts and human review, whereas more capable models can complete a task in one go. Therefore, "cost-per-task" better reflects actual value. Data from the corporate spend management platform Ramp shows that enterprise adoption of both companies has slowed. Anthropic's strongest model, Fable 5, accounts for only 11% of its Claude software spending, indicating that enterprises have a ceiling on AI payments and that higher performance does not necessarily justify a higher price.