Prompt Caching Savings Calculator

What is Prompt Caching and Why You Need It Now

In mid-2024, Anthropic, DeepSeek, and Google revolutionized API pricing by introducing Prompt Caching. Instead of paying full price to repeatedly send the same massive System Prompts (like a 100k-token codebase or RAG documents) to the AI, you can now cache them on the server.

The 90% Discount Rule

When you cache a prompt, models like Claude 3.5 Sonnet and DeepSeek-V3 drop the input token price by up to 90%. For example, sending 1 million input tokens to Claude 3.5 normally costs $3.00. But if those tokens are a “Cache Hit”, the price plummets to just **$0.30**!

How to Read the Calculator Results

• Standard Cost: What you currently pay per month if you send the full context with every single API call without utilizing cache headers.
• Cached Cost: Your new monthly bill assuming the static context stays intact in the server’s cache (usually 5 to 60 minute TTLs depending on the provider).
• Cache Miss Rate: Realistically, the cache resets if inactive. Setting this to 5% means 5 out of 100 API calls will pay full price to write the cache again.