Caching

Caching

Anthropic, OpenAI, and Gemini all cache the stable parts of a request automatically — no setup, applies to every eligible call. Mizan passes the discount straight through, with no markup of its own.

Agentic caching

Mizan routes every wallet-billed call through one shared provider account, so a warm cache carries across your agent's turns, retries, and subagents — even when it was another Mizan customer's call that warmed it seconds earlier. Nobody's actual requests or responses are ever shared, only the provider's own cache infrastructure.

Response Caching

An identical repeat request to the same key is served straight from Mizan's own cache — no provider call, no tokens billed, just the cached response back in milliseconds. Opt-in, configured per key below.