← All news·2026-06-29·3 min read

Amazon engineers are boiling down Anthropic Claude for cost reduction.

Amazon is distilling Anthropic Claude models to create cheaper versions ahead of Anthropic’s new token pricing scheme which will increase costs significantly when their contract comes up for renewal next year.

amazonanthropicllmcostsdistillation

According to sources, Amazon is distilling Anthropic Claude models by training a cheaper model using the outputs of a larger Anthropic model. They are doing this to avoid paying higher fees under Anthropic’s upcoming token pricing scheme which will come into effect once their current contract ends next year.

They are also considering using other models such as those developed by OpenAI or their own Nova family of models. It’s interesting to see one of the biggest investors in the AI space, who has pledged to invest up to $25 billion into Anthropic, trying to hedge their bets.

What distillation means in practice

Model distillation involves using an expensive model to generate training data which can be used to train a smaller, faster model. This produces a model which retains most of the functionality while requiring less compute power and tokens. While this is a perfectly valid approach, it can be tricky to do when dealing with proprietary models which have clauses preventing you from creating similar products.

We’re seeing more and more of these cases popping up in the enterprise AI world. With flagship models still commanding a premium price tag, many organizations are discovering that they can achieve 70–80% of the performance with only 10–15% of the cost via distillation or by using alternative models. This presents an interesting challenge for AI research labs who need to find a way to maintain profitability without losing customers due to price increases or thinning margins.

Source: the-decoder.com

Free course

Stop reading about AI — start building with it

The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.

Start free →
EAEvgenii Arsentev

Author

Evgenii Arsentev

PhD · Founder & CEO, Dad4Kids