Amazon distilling Anthropic models to slash costs ahead of price hikes

amazon engineers are reportedly distilling anthropic's claude models to build proprietary, smaller models before the company shifts from compute-hour to token-based pricing next year. this strategy, confirmed by sources, allows amazon to maintain model performance while reducing inference costs, utilizing rights similar to apple's arrangement with google gemini. although amazon bedrock supports distillation for nova and llama, it excludes anthropic, making this a significant internal workaround. the cost-saving measures coincide with amazon's massive $8b investment in anthropic and $50b in openai, reflecting a broader push to diversify ai infrastructure amid escalating operational expenses.