How AI describes Google Gemini API

The Gemini API provides implicit caching to optimize performance and reduce costs by reusing input tokens across requests. Implicit caching is enabled by default for Gemini 2.5 and newer models and works for both stateful and stateless conversations; the Interactions API does not support explicit caching, and to use explicit caching you must switch to the generateContent API. You can see cache hits in usage.total_cached_tokens, and the minimum input token count for context caching varies by model (e.g., 4096 for Gemini 3.5 Flash / 2048 for Gemini 2.5 Flash).