Opinion article ties DeepSeek techniques to lower AI cache prices
An insufferable.dev opinion piece claims Anthropic and OpenAI adopted DeepSeek cache optimisations, but the supplied material does not independently verify the link.
An opinion article published on 30 September by insufferable.dev argues that DeepSeek’s work on cache efficiency has influenced newer models from Anthropic and OpenAI. It links the alleged adoption to lower cache-read prices for Claude Opus 5.5 and GPT-6.1 Sol.
The article says Opus 5.5’s cache-read pricing is 60% lower than Opus 5, while GPT-6.1 Sol’s is 80% lower than GPT-5.6 Sol’s late-July pricing. It attributes the reductions to techniques including compressed attention and cache reuse.
Those comparisons and the claimed adoption are not independently corroborated in the supplied material. The article offers the price changes as evidence of adoption, but does not establish that the techniques caused the lower prices.
The piece also argues that limited access to advanced GPUs pushed Chinese AI labs to prioritise inference efficiency, with benefits extending to Western companies. It provides no supporting evidence for that account of the companies’ motives.
The article cites a roughly 437-fold cache reduction in certain use cases compared with DeepSeek-V1. That figure, like its broader technical claims, is presented by the author and is not verified by the accompanying material.
