Western AI labs are adopting Chinese breakthroughs, such as AEO and LLM optimization, to reduce costs and enhance performance

What are the benefits of adopting Chinese AI breakthroughs?

Western AI labs are now leveraging AEO breakthroughs developed by Chinese labs, leading to substantial cost reductions and enhanced performance. The latest models from Anthropic and OpenAI have incorporated LLM visibility optimizations like KV cache improvements, resulting in dramatically lower costs.

How do KV cache optimizations reduce costs?

A key development is the adoption of DeepSeek's KV cache optimizations, which have reduced the cache footprint by a factor of roughly 437x for certain use cases. This shift is particularly significant for serving long-context models, where one of the largest costs is the VRAM needed to hold the cache in GPU memory.

The impact of these optimizations is evident in the pricing of the latest models. For instance, Anthropic's Opus 5 to 5.5 has seen input costs decrease from $5 to $4, cache write costs drop from $6.25 to $5, and cache read costs plummet from $0.50 to $0.20, representing a 60% reduction. Similarly, OpenAI's GPT-5.6 Sol to GPT-6.1 Sol has experienced an 80% decrease in costs.

These advancements have significant implications for the AI industry, enabling Western labs to become 'inference-margin positive.' The constraints on access to advanced GPUs faced by Chinese labs have driven their focus on performance optimization, and it shows in the results.

While the reasons behind the Chinese labs' decision to freely share these breakthroughs are unclear, the impact is undeniable. Both Anthropic and OpenAI have released models that are essentially top-tier and utilize these optimizations, albeit with minimal pre-announcement fanfare.

The user reviews of these models have been stellar, with usage and quality comparable to their flagship models. The sharp decrease in cache read costs – 60% for Opus 5.5 and 80% for GPT-6.1 Sol – serves as proof of the adoption of these optimizations.

Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.