Prompt Caching

Evidence (5)

• 6OpenAI Blog·9/22/2026model_release
  GPT-6 Gets Better Prompt Caching

  OpenAI says GPT-6 improves prompt caching by increasing cache hit rates and adding new diagnostics, explicit breakpoints, and controls. The update is intended to reduce latency and lower serving costs for applications using the model.

• 6The New Stack·8/27/2026product_launch
  Anthropic Files API saves time, not token costs

  Anthropic moved its Files API and computer-use toolset out of beta and launched a browser-use toolset. A test comparing file upload, repeated pasting, and prompt caching found that Files API and pasting produced similar token usage and identical answer accuracy, while prompt caching substantially reduced billed input tokens after the initial write.

• 5The New Stack·8/13/2026research
  Why AI pipelines can cost far more after launch

  The article argues that AI application costs often rise sharply in production because token usage grows with prompts, history, retrieval, tool output, and multi-step agent workflows. It says the biggest savings usually come from reducing unnecessary tokens through better architecture, prompt design, caching, routing, and memory management rather than simply switching to a cheaper model.

• 6Hacker News·7/31/2026research
  Why AI session portability is fading

  The article argues that inference APIs are increasingly returning provider-bound state instead of a fully portable transcript. It says features like encrypted reasoning tokens, hidden subagent messages, server-side response IDs, and non-portable search or cache references make AI sessions harder to inspect, export, replay, audit, or delete independently of the provider.

• 6The New Stack·7/23/2026research
  Prompt Caching and Batched Pipelines to Cut RAG Costs

  The article argues that naive retrieval-augmented generation (RAG) architectures often fail in production because synchronous ingestion, timeout limits, and rate limits do not scale to enterprise workloads. It recommends asynchronous uploads, micro-batching, and controlled embedding workers, and discusses prompt caching as one way to reduce costs without hurting accuracy.