← API pricing · prompt caching
Prompt caching prices compared
When part of a prompt repeats from one request to the next, such as a long system prompt or a document, most APIs bill it at a lower cached-input price. Here is that price for each of the 39 live models that publish one, cheapest first, and how much it saves against the normal input price. List prices from the models.dev catalog as of October 25, 2026.
| Model | Cached input | Input | Saving | AW Index |
|---|---|---|---|---|
| MiMo-V2.5Xiaomi | $0.0028 | $0.14 | 98% | — |
| MiMo-V2.6 ProXiaomi | $0.0036 | $0.435 | 99% | 65.1 |
| Claude Haiku 5.5Anthropic | $0.01 | $0.10 | 90% | 50.5 |
| GPT-6 LunaOpenAI | $0.01 | $0.10 | 90% | 28 |
| MiniMax M3.1 FlashMiniMax | $0.015 | $0.15 | 90% | 48.8 |
| Qwen 3.8 FlashAlibaba | $0.016 | $0.15 | 89% | — |
| DeepSeek V4 ProDeepSeek | $0.022 | $0.66 | 97% | 31.8 |
| Gemini 3.5 Flash-LiteGoogle | $0.03 | $0.30 | 90% | — |
| GLM-5.3 FlashZhipu AI | $0.03 | $0.15 | 80% | 32.4 |
| MiniMax M3MiniMax | $0.06 | $0.30 | 80% | 5.8 |
| Gemini 3.6 FlashGoogle | $0.075 | $0.75 | 90% | — |
| Gemini 3.7 FlashGoogle | $0.075 | $0.75 | 90% | — |
| Gemini 3.8 FlashGoogle | $0.075 | $0.75 | 90% | 43.3 |
| Claude Sonnet 5.5Anthropic | $0.10 | $2 | 95% | 93.5 |
| Gemini 4 ArgonGoogle | $0.10 | $2 | 95% | 79.5 |
| GPT-6.1 SolOpenAI | $0.10 | $2 | 95% | 82.9 |
| Mistral Large 4Mistral | $0.14 | $1.36 | 90% | 32.6 |
| Gemini 3.5 FlashGoogle | $0.15 | $1.50 | 90% | 21.3 |
| Muse Spark 1.1Meta | $0.15 | $1.25 | 88% | 30.8 |
| Muse Spark 1.2Meta | $0.15 | $1.25 | 88% | — |
| Muse Spark 1.3Meta | $0.15 | $1.25 | 88% | 74.9 |
| Claude Opus 5.5Anthropic | $0.20 | $4 | 95% | 100 |
| Claude Sonnet 5Anthropic | $0.20 | $2 | 90% | 27.9 |
| GPT-6 SolOpenAI | $0.20 | $2 | 90% | 68.6 |
| Kimi K3.1 ProMoonshot | $0.20 | $2 | 90% | 86 |
| Fable 5.1Anthropic | $0.25 | $10 | 98% | 90.7 |
| Qwen 3.8 MaxAlibaba | $0.25 | $2 | 88% | 62.9 |
| GLM-5.2Zhipu AI | $0.26 | $1.40 | 81% | 21.2 |
| GLM-5.3Zhipu AI | $0.26 | $1.40 | 81% | 60 |
| Kimi K3Moonshot | $0.30 | $3 | 90% | 27.3 |
| Grok 4.5xAI | $0.30 | $2 | 85% | 34.9 |
| ChatGPT-5.6OpenAI | $0.40 | $4 | 90% | 50 |
| Claude Opus 4.8Anthropic | $0.50 | $5 | 90% | 37.2 |
| Claude Opus 5Anthropic | $0.50 | $5 | 90% | 72.8 |
| GPT-5.5OpenAI | $0.50 | $5 | 90% | 42.6 |
| Grok 4.6xAI | $0.50 | $2 | 75% | 58.1 |
| Grok 4.7xAI | $0.50 | $2 | 75% | 59.4 |
| Fable 5Anthropic | $1 | $10 | 90% | 72.1 |
| GPT-6 AstraOpenAI | $1 | $10 | 90% | 87.4 |
Prices per million tokens. Saving is the cached price's discount on the same model's normal input price.
To see what caching does to a real bill, set the cached share in the cost calculator (it opens at half the input from cache).
Quick answers
Which LLM API has the cheapest cached input?
MiMo-V2.5, at $0.0028 per million cached input tokens, the lowest of the 39 models that publish a cached price.
Which model gives the biggest prompt-caching discount?
MiMo-V2.6 Pro: cached input costs 99% less than normal input ($0.0036 against $0.435 per million tokens).
How much does prompt caching save?
Across the 39 models, cached input costs between 75% and 99% less than normal input; the middle model saves 90%. It only applies to the part of a prompt that repeats, such as a long system prompt or a document sent with every request.
What does this comparison leave out?
Some makers also charge to write a prompt into the cache or to keep it stored, and set how long a cached prompt lasts. The models.dev catalog lists the cached read price, which is what this page compares, as of October 25, 2026.
Every price in one table: LLM API pricing. For an alert when a new model goes live, watch free.
Cite this page
Writing about prompt caching prices? Copy a link or a citation.
<a href="https://artificialwatch.com/llm-pricing/prompt-caching">Prompt caching prices compared</a>ArtificialWatch, "Prompt caching prices compared", https://artificialwatch.com/llm-pricing/prompt-cachingPrices move. Get an alert when a new model lands.
Free: a Chrome push and an email about 15 minutes after a new model answers on a public API. Members get a text or a phone call the moment a launch is confirmed.
- Free forever
- No card
- Unsubscribe in one click