Anthropic is shifting its consumer AI model pricing strategy, introducing usage-based fees for access to Claude Fable 5 starting July 12. Users of the company’s $20, $100, and $200-a-month plans will now pay additional charges for using the model, marking the first time a frontier AI lab has implemented this approach for consumer access. The new pricing structure mirrors the company’s API rates, with $10 per million tokens sent to the model and $50 per million tokens generated to answer user questions. For example, a user on the $20-a-month plan who sends and receives a million tokens in July would owe an extra $60, bringing their total monthly cost to $80. This amount is comparable to about five months of Amazon Prime membership.
The new pricing model reflects the growing complexity of AI usage, as newer models like Fable 5 can consume more tokens in internal reasoning processes. While usage-based billing has been standard for developers accessing models through APIs, Anthropic has historically relied on flat monthly subscriptions for consumer revenue. However, the company is aligning with industry trends that favor usage-based pricing, following similar moves by AI startups like Cursor and shifting to usage-based billing for large business customers. Anthropic’s spokesperson, Reem Ateyeh, noted the company aims to return Fable 5 to subscription plans when sufficient capacity allows, though data center constraints remain a challenge.
Anthropic’s decision to implement usage-based fees for Fable 5 comes after an extended promotional period with no additional charges. The company initially anticipated high demand for the model, which has seen significant growth since its general release on July 1. With 245 million unique visitors in May, Claude is gaining traction in the consumer space, though it still lags behind ChatGPT and Google’s Gemini in terms of user numbers. The pricing change is a test of consumer willingness to pay for premium AI models, as Anthropic positions itself as a competitor to OpenAI and Google in the AI landscape.
Source: wired
