OpenAI and Anthropic Shift AI Pricing from Per-Token to Per-Task Costs Amid Low-Price Competition

OpenAI and Anthropic are pushing AI pricing standards from "per-token cost" to "per-task cost," arguing that high-premium models can complete tasks efficiently in a single pass, thereby lowering overall costs. Both companies have made statements to this effect over the past few weeks. Jonathan Pelosi, Anthropic's head of financial services, noted in a recent interview that token cost is merely a proxy metric for how much AI compute is consumed; for enterprises, a better proxy is the actual cost of completing a task.

The shift in pricing narrative comes against a backdrop of multiple vendors squeezing the market with extremely low token prices. DeepSeek, for example, launched its flagship open-source model in April with input and output token costs that are a fraction of those of Anthropic's most advanced products. Companies that once encouraged employees to maximize AI usage have now set limits after being hit by bill shocks.

Third-party data offers a more direct signal. Anthropic's strongest and most expensive widely available model, Fable 5, accounts for only 11% of its Claude software spending. On a per-task cost basis, Anthropic and OpenAI models remain the most expensive on the market, with Anthropic's Claude Fable 5 topping the list; however, both companies also offer more competitive low-priced products, such as OpenAI's GPT-5.6 Luna. A growing number of customers are turning to third-party benchmark institutions such as Artificial Analysis and Vals AI for independent evaluations to compare models on a like-for-like basis.