IonBin
All pages
Search
Concept

Prompt caching

Prompt caching reuses processing of eligible repeated input, such as instructions or reference documents. Cache hits can reduce input cost and latency, but minimum lengths, configuration, write charges and retention differ by model. Put stable content first and verify actual cache usage.

Verified Oct 18, 2026 Review by Apr 16, 2027 4 sources

What it is

Prompt caching stores the model’s processing of a prompt prefix, so later requests with the same prefix skip that work. OpenAI, Anthropic and Google offer it with different controls. OpenAI supports explicit breakpoints on GPT-5.6 and later; earlier models use implicit caching. Anthropic requires cache_control configuration for its automatic or explicit modes. Google enables implicit caching by default on Gemini 2.5 and newer.

How it works

  1. Put the unchanging part of the prompt first: instructions, examples, documents.
  2. The first request processes and caches that prefix.
  3. Later eligible requests may reuse that prefix. Inspect the usage fields for actual cache hits rather than assuming every repeated prompt was cached.

Current terms (checked 2026-10-18):

  • Anthropic. Read multipliers vary: Sonnet 5.5 and Opus 5.5 use 0.05× base input, Fable 5.1 uses 0.025×, and Haiku 5.5 uses 0.1×. Writes cost 1.25× for a 5-minute cache or 2× for a 1-hour cache.
  • OpenAI. On GPT-5.6 and later, cache writes cost 1.25× the uncached input rate, the minimum cacheable prefix is 1,024 tokens, and the minimum default lifetime is 30 minutes from the latest write or reuse. Reads use 0.1×, except GPT-6.1 Sol at 0.05×.
  • Google. Each Gemini model has a minimum prompt size. The Interactions API and Gemini 4 Argon onward support implicit caching only; explicit caching remains available for supported older models through generateContent.

Benefits

  • Potentially lower input cost for chatbots, agents and document Q&A that reuse context.
  • Less repeated prompt processing can reduce latency.
  • Makes long, detailed system prompts more affordable.

Limits and risks

  • Only exact prefix matches count. A change early in the prompt can prevent reuse of the following content; an earlier matching segment may still qualify.
  • Caches expire, and writes can cost more than normal input. Short or rarely reused prompts may not save money.
  • Rates, lifetimes and minimums differ by vendor and model, so re-check before you budget.

Checked specs

Snapshot from official sources on Oct 18, 2026.

OpenAI GPT-5.6 and later1.25× writes; 1,024 visible-token minimum; at least 30 minutes from latest write/reuse. Read multiplier is model-specific. [1]
AnthropicWrite and read rates differ by model and lifetime; configure cache_control, then inspect actual usage. [2]
Google GeminiImplicit caching enabled by default on Gemini 2.5+; newer API/model paths do not support explicit caches. [4]

Best-fit stacks

Stack 08

Dev, web & IT

Design, code, host and ship sites and internal tools, with an AI coding agent doing most of the typing.

Agent compatibility

Not tool-specific. See Agent-ready for connector support by tool.

Levels come from each tool's recorded, sourced compatibility. How the levels work.

Sources

  1. Prompt caching · OpenAIofficial accessed Oct 18, 2026
  2. Prompt caching · Anthropicofficial accessed Oct 18, 2026
  3. Models & pricing · DeepSeekofficial accessed Oct 18, 2026
  4. Context caching · Googleofficial accessed Oct 18, 2026

Practitioner views are restated in our words and labeled as views, not measured results. Product facts come from official sources dated above.

Suggest an edit

IonBin's wiki isn't user-edited. Corrections go to our Directory Curator, who checks them against primary sources before anything changes. Spot something wrong or out of date on “Prompt caching”? Email us with a link to a primary source.

Suggest an edit by email or write to contact@ionbin.com