Amazon engineers are reportedly distilling Anthropic models to create smaller, cheaper versions for internal use ahead of a new token-based pricing structure. This strategy follows a report from The Information regarding concerns over rising expenses. Distillation involves training a smaller model to replicate the outputs of a larger one, a process Amazon is permitted to perform under its current partnership terms. While the company offers a distillation service on its Bedrock cloud platform, it currently supports only its own Nova models and Meta’s Llama models, excluding Claude. The effort stems from ongoing negotiations about how Amazon pays for access to these tools. Starting next year, fees will shift from compute hours to tokens processed, a change that could significantly increase costs for the retailer. An Amazon spokesperson stated that the expanded partnership will not raise prices, while Anthropic highlights the value provided relative to model performance. Amazon is also reportedly exploring alternatives such as OpenAI and its internal Nova models. The company has invested up to $25 billion more in Anthropic and up to $50 billion in OpenAI this year.
The move highlights a strategic shift toward cost control as usage-based pricing becomes standard. Amazon seeks to reduce reliance on external models for internal tasks where performance requirements are less demanding. This approach allows the company to maintain efficiency while negotiating better terms for high-stakes applications. The competition between cloud providers and AI firms intensifies as pricing models evolve.
* Amazon holds rights to distill models despite excluding them from its public Bedrock platform
* Pricing will shift from compute hours to tokens processed starting next year
* Investment totals include $25 billion for Anthropic and $50 billion for OpenAI this year
