How prompt caching bills a repeated prompt start

OpenAI's prompt caching guide says prompt caching reuses work when requests share the same prompt prefix. Quote: "Prompt caching reuses work when requests share the same prompt prefix."

Prompt caching reuses work when requests share the same prompt prefix.

OpenAI says the prompt cache stores key-value (KV) tensors, not the tokens themselves, and describes KV states as intermediate states the model calculates while processing input tokens. Quote: "The prompt cache stores key-value (KV) tensors, not the tokens themselves."

When the model processes input tokens, it must calculate intermediate states, known as key-value (KV) states. These states let the model refer back to earlier tokens while processing new input and generating output tokens. Prompt caching preserves that state for a reusable prefix : the unchanged tokens at the beginning of a prompt. When a later request has the same prefix and finds a matching cache entry, the model can reuse the saved state instead of processing those tokens again. It still needs to process any new input to generate a new response. The prompt cache stores key-value (KV) tensors, not the tokens themselves.

OpenAI says that when a later request has the same prefix and finds a matching cache entry the model can reuse the saved state, but it still needs to process any new input. Quote: "It still needs to process any new input to generate a new response."

When a later request has the same prefix and finds a matching cache entry, the model can reuse the saved state instead of processing those tokens again. It still needs to process any new input to generate a new response.

OpenAI says cache reuse requires the entire rendered prefix to match. Quote: "Cache reuse requires the entire rendered prefix to match."

Cache reuse requires the entire rendered prefix to match. If content or a relevant setting changes before a breakpoint, the prefix after that change cannot match the existing cache entry.

OpenAI's list of settings that affect the cached prefix includes model, tools, text.format and reasoning.effort. Quote: "Which settings affect the cached prefix?"

The main settings to check are: Setting Impact model A different model can use different weights and caching behavior. tools Changes tool names, descriptions, schemas, ordering, or tool-specific instructions. parallel_tool_calls Can change instructions about calling multiple tools in one turn. text.format ( Structured Outputs ) Adds output-format instructions and the requested schema. reasoning.effort Can change model-side reasoning instructions.

OpenAI says the minimum cacheable prompt length is 1,024 tokens for GPT-5.6 and later. Quote: "The minimum cacheable prompt length is 1,024 tokens for GPT‑5.6 and later"

The minimum cacheable prompt length is 1,024 tokens for GPT-5.6 and later and varies by request settings for earlier models.

OpenAI says prompt caching is enabled by default for supported OpenAI models. Quote: "Prompt caching is enabled by default for supported OpenAI models."

Prompt caching is enabled by default for supported OpenAI models.

OpenAI says that for GPT-5.6 and later both implicit and explicit caching are supported. Quote: "Both implicit and explicit caching are supported, where explicit caching gives you more control over which context is written to cache."

Both implicit and explicit caching are supported, where explicit caching gives you more control over which context is written to cache.

OpenAI says that for GPT-5.6 and later a cached prefix remains eligible for reuse for 30 minutes after its most recent write or reuse. Quote: "A cached prefix remains eligible for reuse for 30 minutes after its most recent write or reuse, though OpenAI may retain it longer."

A cached prefix remains eligible for reuse for 30 minutes after its most recent write or reuse, though OpenAI may retain it longer.

OpenAI says that for GPT-5.6 and later cache writes cost 1.25 times the standard uncached input-token rate and reads cost 0.1 times on most of these models and 0.05 times on GPT-6.1 Sol. Quote: "cache writes cost 1.25× the standard, uncached input-token rate. Subsequent reads cost 0.1× that rate on most of these models and 0.05× on"

For GPT-5.6 and later, cache writes cost 1.25× the standard, uncached input-token rate. Subsequent reads cost 0.1× that rate on most of these models and 0.05× on GPT-6.1 Sol .

OpenAI says that across ten requests, one write and nine full reads cost 2.15 times at the 0.1 times read rate, compared with 10 times without caching. Quote: "Across ten requests, one write and nine full reads cost 2.15× at that rate, compared with 10× without caching."

Across ten requests, one write and nine full reads cost 2.15× at that rate, compared with 10× without caching.

OpenAI says to track usage.input_tokens_details.cached_tokens and usage.input_tokens_details.cache_write_tokens to measure actual cache performance. Quote: "Track usage.input_tokens_details.cached_tokens, usage.input_tokens_details.cache_write_tokens"

Track usage.input_tokens_details.cached_tokens , usage.input_tokens_details.cache_write_tokens , input-token counts, latency, and realized cost.

Anthropic says cache hits require 100% identical prompt segments, including all text and images up to and including the block marked with cache control. Quote: "Cache hits require 100% identical prompt segments, including all text and images up to and including the block marked with cache control."

Exact matching: Cache hits require 100% identical prompt segments, including all text and images up to and including the block marked with cache control.

Anthropic says prompt caching references the entire prompt, tools, system and messages in that order, up to and including the block designated with cache_control. Quote: "Prompt caching references the entire prompt:"

Prompt caching references the entire prompt: tools , system , and messages (in that order), up to and including the block designated with cache_control .

Anthropic says changes at each level of the tools, system, messages hierarchy invalidate that level and all subsequent levels. Quote: "Changes at each level invalidate that level and all subsequent levels."

As described in Structuring your prompt , the cache follows the hierarchy: tools → system → messages . Changes at each level invalidate that level and all subsequent levels.

Anthropic says modifying tool definitions (names, descriptions, parameters) invalidates the entire cache. Quote: "Modifying tool definitions (names, descriptions, parameters) invalidates the entire cache"

Tool definitions ✘ ✘ ✘ Modifying tool definitions (names, descriptions, parameters) invalidates the entire cache

Anthropic lists a minimum cacheable prompt length of 512 tokens for Claude Fable 5.1, Mythos 5.1, Opus 5.5, Opus 5, Sonnet 5.5, Fable 5, Mythos 5 and Haiku 5.5. Quote: "512 tokens for Claude Fable 5.1"

512 tokens for Claude Fable 5.1, Claude Mythos 5.1 , Claude Opus 5.5, Claude Opus 5, Claude Sonnet 5.5, Claude Fable 5, Claude Mythos 5 , and Claude Haiku 5.5

Anthropic says shorter prompts cannot be cached, even if marked with cache_control. Quote: "Shorter prompts cannot be cached, even if marked with"

Shorter prompts cannot be cached, even if marked with cache_control . Any requests to cache fewer than this number of tokens will be processed without caching, and no error is returned.

Anthropic says automatic caching, a single cache_control field at the top level of the request, applies the cache breakpoint to the last cacheable block and moves it forward as conversations grow. Quote: "The system automatically applies the cache breakpoint to the last cacheable block and moves it forward as conversations grow."

Automatic caching : Add a single cache_control field at the top level of your request. The system automatically applies the cache breakpoint to the last cacheable block and moves it forward as conversations grow.

Anthropic says a prompt can define up to 4 cache breakpoints. Quote: "You can define up to 4 cache breakpoints if you want to:"

You can define up to 4 cache breakpoints if you want to:

Anthropic says the cache has a 5-minute lifetime by default and is refreshed for no additional cost each time the cached content is used. Quote: "By default, the cache has a 5-minute lifetime. The cache is refreshed for no additional cost each time the cached content is used."

By default, the cache has a 5-minute lifetime. The cache is refreshed for no additional cost each time the cached content is used.

Anthropic says it also offers a 1-hour cache duration at additional cost. Quote: "Anthropic also offers a 1-hour cache duration"

If you find that 5 minutes is too short, Anthropic also offers a 1-hour cache duration at additional cost .

Anthropic says 5-minute cache write tokens are 1.25 times the base input price, 1-hour cache write tokens are 2 times, and cache read tokens are 0.1 times, with per-model exceptions in the table footnote. Quote: "5-minute cache write tokens are 1.25 times the base input tokens price"

5-minute cache write tokens are 1.25 times the base input tokens price 1-hour cache write tokens are 2 times the base input tokens price Cache read tokens are 0.1 times the base input tokens price (see the table footnote for per-model exceptions)

Anthropic says cache hits and refreshes on Claude Opus 5.5 and Claude Sonnet 5.5 are priced at 0.05x the base input price. Quote: "Cache hits and refreshes on Claude Opus 5.5 and Claude Sonnet 5.5 are priced at 0.05x the base input price."

Cache hits and refreshes on Claude Opus 5.5 and Claude Sonnet 5.5 are priced at 0.05x the base input price.

Anthropic names the response usage fields cache_creation_input_tokens, cache_read_input_tokens and input_tokens. Quote: "cache_creation_input_tokens: Number of tokens written to the cache when creating a new entry."

cache_creation_input_tokens : Number of tokens written to the cache when creating a new entry. cache_read_input_tokens : Number of tokens retrieved from the cache for this request. input_tokens : Number of input tokens which were not read from or used to create a cache

Anthropic says prompt caching has no effect on output token generation and that the response is identical to what you would get without caching. Quote: "Prompt caching has no effect on output token generation. The response you receive is identical to what you would get if prompt caching were not used."

Output token generation: Prompt caching has no effect on output token generation. The response you receive is identical to what you would get if prompt caching were not used.