Prompt Caching ROI Calculator

Prompt caching stores frequently repeated parts of your prompt (like system prompts) in the model's compute cache. On subsequent requests, the cached portion is processed at a fraction of the normal cost — up to 90% cheaper on Anthropic Claude.

Anthropic Claude offers 90% off cached tokens (cache reads cost 10% of normal). OpenAI offers 50% off cached tokens for GPT-4o and newer models. Google Gemini also supports context caching for long documents.

On Anthropic, the cache lasts at least 5 minutes per cache entry, with TTL resetting on each hit. For high-traffic applications, caches effectively persist indefinitely. OpenAI's cached tokens are based on prefix matching with no explicit TTL.

Writing a new cache entry costs 25% more than a regular input token on Anthropic (1.25× normal price). This one-time write cost is quickly offset by cheap cache reads on subsequent requests.

For Anthropic, add cache_control: { type: 'ephemeral' } to the content blocks you want cached. For OpenAI, caching happens automatically for matching prefixes over 1,024 tokens. Digiqt can help you implement caching in your existing stack.