OpenAI and Anthropic Push Cost Per Task AI Metric

AI developers OpenAI and Anthropic are advocating for a new industry metric called cost per task to replace the current standard of cost per token. The companies argue that token-based pricing is an imprecise proxy for business expenses because cheaper models often require more attempts or human review to achieve a successful result.

Sarah Friar, Chief Financial Officer of OpenAI, stated that while lower-cost models may have cheaper tokens, more capable models can often complete the same task in a single pass. This shift in narrative comes as both firms face pricing pressure from Meta, SpaceX, and Chinese competitors such as DeepSeek, which offer lower token costs.

Despite the push for this new metric, high-end models remain expensive. While OpenAI's GPT-5.6 Luna is considered affordable under the task-based metric, Anthropic's Claude Fable 5 remains among the most costly options. Data from Ramp indicates that business adoption of these premium models has slowed as customers increasingly migrate toward cheaper open-source alternatives.