Prompt Caching is an optimization technique that stores prefix token representations of long prompts in memory, allowing subsequent API queries with the same prefix to reuse states, reducing latency and cost.
Directly governs the hardware efficiency and hardware-level token throughput when deploying long conversation context management, document question-answering systems, and agent tool setups; optimizing Prompt Cache is a major factor in compute cost budgeting.
Prompt Caching is an API-level optimization that caches the key-value states of long prompt prefixes in memory. When a new user request shares the same prefix (such as system prompts, documents, or early chat history), the server reuses the cached states, reducing pre-fill compute requirements, latency, and costs.
API providers charge significantly less for cached tokens because they do not require GPU computations for pre-filling attention states.
Static system instructions, uploaded document context, or early conversation turns that are shared across requests.
Reference this definition in your articles, research, or documentation to credit this source:
OpenAI GPT-5.6 Sol, Terra, and Luna are now generally available on Amazon Bedrock, along with explicit prompt caching that gives you precise control over.