Pricing & Cost

Cached Tokens

Prompt caching stores previously processed tokens, reusing KV cache. Cached tokens cost 10-25% of standard token price. Caching is valuable when using common context repeatedly (RAG, system prompts). Minimum context size required before caching helps (usually 1K tokens). Caching works best for static context (documents) with varying queries. Prompt caching is increasingly available in modern APIs. Caching can dramatically reduce costs for some applications.