Prompt caching savings calculator
What caching the repeated part of your prompt changes on each model, from its size and your number of calls.
Result
On Claude Opus 5.5, caching takes these 100 calls from $28.20 to $5.69: you save $22.51 (80%).
| Model | Without cache | With cache | Saved | Saved % |
|---|---|---|---|---|
| MiMo-V2.6-ProXiaomi | $2.87 | $0.31 | $2.56 | 89% |
| DeepSeek V4 ProDeepSeek | $4.46 | $0.67 | $3.79 | 85% |
| DeepSeek V4.1 FlashDeepSeek | $2.07 | $0.32 | $1.75 | 84% |
| Claude Fable 5.1Anthropic | $70.50 | $12.73 | $57.77 | 82% |
| GPT-6.1 SolOpenAI | $14.10 | $2.81 | $11.29 | 80% |
| Gemini 4 ArgonGoogle | $14.10 | $2.81 | $11.29 | 80% |
| Muse WonderMeta | $28.20 | $5.63 | $22.57 | 80% |
| Claude Opus 5.5Anthropic | $28.20 | $5.69 | $22.51 | 80% |
| Claude Sonnet 5.5Anthropic | $14.10 | $2.84 | $11.26 | 80% |
| Mistral Large 4Mistral | $9.20 | $1.95 | $7.25 | 79% |
| MiniMax M3.1 FlashMiniMax | $1.03 | $0.23 | $0.80 | 77% |
| Qwen3.8 MaxAlibaba (Qwen) | $13.50 | $3.10 | $10.39 | 77% |
| Muse Spark OpenMeta | $8.51 | $1.98 | $6.53 | 77% |
| GPT-6 AstraOpenAI | $70.50 | $17.04 | $53.46 | 76% |
| GPT-6 SolOpenAI | $14.10 | $3.41 | $10.69 | 76% |
| GPT-6 LunaOpenAI | $0.70 | $0.17 | $0.53 | 76% |
| GPT-5.6 SolOpenAI | $28.20 | $6.82 | $21.38 | 76% |
| Gemini 3.8 FlashGoogle | $5.29 | $1.28 | $4.01 | 76% |
| Gemini 3.6 FlashGoogle | $5.29 | $1.28 | $4.01 | 76% |
| Kimi K3Moonshot AI | $21.15 | $5.11 | $16.04 | 76% |
| Claude Fable 5Anthropic | $70.50 | $17.19 | $53.31 | 76% |
| Claude Opus 5Anthropic | $35.25 | $8.60 | $26.66 | 76% |
| Claude Opus 4.8Anthropic | $35.25 | $8.60 | $26.66 | 76% |
| Claude Opus 4.7Anthropic | $35.25 | $8.60 | $26.66 | 76% |
| Claude Opus 4.6Anthropic | $35.25 | $8.60 | $26.66 | 76% |
| Claude Sonnet 5Anthropic | $14.10 | $3.44 | $10.66 | 76% |
| Claude Sonnet 4.6Anthropic | $21.15 | $5.16 | $15.99 | 76% |
| Claude Haiku 4.5Anthropic | $7.05 | $1.72 | $5.33 | 76% |
| Claude Haiku 5.5Anthropic | $0.70 | $0.17 | $0.53 | 76% |
| Grok 4.3xAI | $8.25 | $2.01 | $6.24 | 76% |
| GPT-5.5OpenAI | $36 | $9.27 | $26.73 | 74% |
| Gemini 3.1 Pro PreviewGoogle | $14.40 | $3.71 | $10.69 | 74% |
| GLM-5.3Z.ai | $9.48 | $2.71 | $6.77 | 71% |
| Kimi K2.6Moonshot AI | $6.58 | $1.89 | $4.69 | 71% |
| Gemini 3.5 Flash-LiteGoogle | $2.27 | $0.66 | $1.60 | 71% |
| GLM-5.3 FlashZ.ai | $1.02 | $0.31 | $0.71 | 70% |
| MiniMax M3MiniMax | $2.07 | $0.64 | $1.43 | 69% |
| Grok 4.7xAI | $13.50 | $4.59 | $8.91 | 66% |
| Grok 4.6xAI | $13.50 | $4.59 | $8.91 | 66% |
| Qwen3.8 27BAlibaba (Qwen) | $3.06 | $1.04 | $2.02 | 66% |
| gpt-oss-120bOpenAI | $1.03 | $0.59 | $0.45 | 43% |
| Mistral Medium 3.5Mistral | $10.57 | n/a | n/a | n/a |
How we work it out
Without caching every call pays the input price for the whole prompt. With caching the first call writes the prefix (Anthropic charges 1.25× the input price for a 5-minute cache[1]), and every later call reads it at the cache price. New tokens and output cost the same either way.
Questions
How does prompt caching save money?
The part of the prompt that repeats on every call (instructions, tool definitions, files) is stored after the first call and read back at a much lower price on the next ones.
Is the first call more expensive?
On Claude, writing the 5-minute cache costs 1.25 times the input price. OpenAI and Google bill the first call at the normal price; Google also charges hourly storage for explicit caches, which this calculator leaves out.
When does caching not help?
When the repeated part is small, or calls are so far apart that the cache expires between them.
More tools
- Compare AI modelsEvery pair, side by side
- AI token cost calculatorWhat your tokens cost on every model
- Context window calculatorWill my code fit in the context window?
- Claude plan vs API calculatorSubscription or pay-as-you-go?
- Free limit reset trackerIs a free Claude or Codex reset live right now?
- Claude usage limit trackerDid the 5-hour or weekly limit go up or down?
- Claude Code usage dashboardYour tokens, models, cost and busiest hours, from one command
Sources
Numbers from the makers' own pages, checked by hand, last read . List prices in US dollars before tax. Spotted a mistake? office@shuriken.sh.
Made by Shuriken, the harness for coding agents. Its hot swapping moves your chats to another of your accounts or API keys when one nears its limit. Use only accounts you own and are entitled to use.
Try Shuriken on your own computerYou're signed in
Add your Claude, Codex and Gemini accounts once. Shuriken switches between them so your agents keep working.Your accounts, projects and chats are in the app.
Coming soon