Anthropic Extends Claude Prompt Caching to 1-Hour TTL

When you use an AI API in a real app, you often send the same big chunk of text over and over — think a long system prompt, a legal document, or a whole codebase. Without caching, you pay to process that text every single time. Anthropic introduced prompt caching earlier, but it had a short time window and didn't work well across a multi-turn conversation. This update fixes both. The cache now lasts a full hour, which is long enough for most user sessions and background jobs. It also supports multi-turn caching, meaning the same cached context can serve multiple back-and-forth exchanges without being re-processed. The cost saving is dramatic: cached tokens are billed at about 10% of the normal rate, a 90% reduction. For enterprise apps that feed long documents or system prompts on every call, this can translate to massive monthly bill reductions. It also speeds up responses because the model doesn't have to re-read the same context from scratch.