Prompt caching pricing compared

84 models charge less for input they have seen before. When the start of your prompt repeats across requests, such as a long system prompt or a document you ask several questions about, those tokens cost typically 90% less.

Prices updated November 14, 2026

Cached input prices per million tokens

ModelProviderInput / 1MCached input / 1MDiscount
deepseek-flashDeepSeek$0.3$0.00698%
deepseek-v4-flashDeepSeek$0.3$0.00698%
deepseek-v4-1-flash-1017Mistral$0.3$0.00798%
claude-fable-5-2Anthropic$8$0.298%
claude-mythos-5-2Anthropic$8$0.298%
claude-fable-5-1Anthropic$10$0.2598%
claude-mythos-5-1Anthropic$10$0.2598%
deepseek-v4-proDeepSeek$1.32$0.04497%
gpt-6.1-terraOpenAI$1$0.0595%
gemini-4-hydrogenGoogle Gemini$1.5$0.07595%
gpt-6.1-solOpenAI$2$0.195%
claude-sonnet-5-5Anthropic$2$0.195%
gemini-4-argonGoogle Gemini$2$0.195%
gemini-4-carbonGoogle Gemini$2$0.195%
claude-opus-5-5Anthropic$4$0.295%
gpt-5-nanoOpenAI$0.05$0.00590%
gpt-6-lunaOpenAI$0.1$0.0190%
claude-haiku-5-5Anthropic$0.1$0.0190%
gemini-2.5-flash-liteGoogle Gemini$0.1$0.0190%
ministral-3-3b-2512Mistral$0.1$0.0190%
ministral-3b-2512Mistral$0.1$0.0190%
mistral-smallMistral$0.1$0.0190%
deepseek-coderDeepSeek$0.14$0.01490%
ministral-3-8b-2512Mistral$0.15$0.01590%
ministral-8b-2512Mistral$0.15$0.01590%
gpt-5.4-nanoOpenAI$0.2$0.0290%
gpt-5.6-lunaOpenAI$0.2$0.0290%
ministral-3-14b-2512Mistral$0.2$0.0290%
ministral-14b-2512Mistral$0.2$0.0290%
gpt-5-miniOpenAI$0.25$0.02590%
gemini-3.1-flash-liteGoogle Gemini$0.25$0.02590%
mistral-tinyMistral$0.25$0.02590%
deepseek-v3.2DeepSeek$0.28$0.02890%
gemini-2.5-flashGoogle Gemini$0.3$0.0390%
gemini-3.5-flash-liteGoogle Gemini$0.3$0.0390%
codestral-2508Mistral$0.3$0.0390%
open-mistral-nemoMistral$0.3$0.0390%
gemini-3-flash-previewGoogle Gemini$0.5$0.0590%
mistral-large-3Mistral$0.5$0.0590%
mistral-large-2512Mistral$0.5$0.0590%
mistral-large-4Mistral$0.7$0.0790%
gpt-5.4-miniOpenAI$0.75$0.07590%
gemini-3.6-flashGoogle Gemini$0.75$0.07590%
gemini-3.7-flashGoogle Gemini$0.75$0.07590%
gemini-3.8-flashGoogle Gemini$0.75$0.07590%
claude-haiku-4-5Anthropic$1$0.190%
gpt-5OpenAI$1.25$0.12590%
gpt-5.1OpenAI$1.25$0.12590%
gemini-2.5-proGoogle Gemini$1.25$0.12590%
gemini-3.5-flashGoogle Gemini$1.5$0.1590%
mistral-mediumMistral$1.5$0.1590%
gpt-5.2OpenAI$1.75$0.17590%
gpt-5.6-terraOpenAI$2$0.290%
gpt-6-solOpenAI$2$0.290%
claude-sonnet-5Anthropic$2$0.290%
gemini-3.1-pro-previewGoogle Gemini$2$0.290%
gpt-5.4OpenAI$2.5$0.2590%
claude-sonnet-4-5Anthropic$3$0.390%
claude-sonnet-4-6Anthropic$3$0.390%
gpt-5.6OpenAI$4$0.490%
gpt-5.5OpenAI$5$0.590%
claude-opus-4-5Anthropic$5$0.590%
claude-opus-4-6Anthropic$5$0.590%
claude-opus-4-7Anthropic$5$0.590%
claude-opus-4-8Anthropic$5$0.590%
claude-opus-5Anthropic$5$0.590%
gpt-6-astraOpenAI$10$190%
gpt-6-astra-lawOpenAI$10$190%
claude-fable-5Anthropic$10$190%
claude-mythos-5Anthropic$10$190%
grok-4.5xAI$2$0.385%
grok-4.3xAI$1.25$0.284%
grok-4.20xAI$1.25$0.284%
grok-build-0.1xAI$1$0.280%
grok-code-fastxAI$1$0.280%
gpt-4.1-miniOpenAI$0.4$0.175%
gpt-4.1OpenAI$2$0.575%
o3OpenAI$2$0.575%
grok-4.6xAI$2$0.575%
grok-4.7xAI$2$0.575%
deepseek-r1DeepSeek$0.55$0.1475%
deepseek-v3DeepSeek$0.27$0.0774%
gpt-4o-miniOpenAI$0.15$0.07550%
gpt-4oOpenAI$2.5$1.2550%

Some providers also charge for writing to the cache or for how long it is kept; check the provider's pricing page for those. To cache, put the part that repeats (instructions, examples, documents) at the start of the prompt.

Calculate with cached input →

Questions and answers

What is prompt caching?

When many requests start with the same text, the provider can keep that part of the prompt processed and reuse it. You pay a lower cached-input price for those tokens.

How much does prompt caching save?

84 models charge less for input they have seen before. When the start of your prompt repeats across requests, such as a long system prompt or a document you ask several questions about, those tokens cost typically 90% less.