Question 1

How accurate is the prompt caching calculator?

Accepted Answer

Token estimates use chars/4 heuristic, accurate to within 5-15% for English prose. Use the Tokens input mode for exact values from your tokenizer.

Question 2

Why does Anthropic show a write cost but OpenAI does not?

Accepted Answer

Anthropic uses explicit cache_control which charges 1.25x base input rate to write the cache entry. OpenAI auto-caches without a separate write fee, charging 0.5x base for cached reads.

Question 3

What is the break-even point?

Accepted Answer

For vendors with a write cost, the call count where per-call savings repay the write cost. Below break-even, caching is more expensive than not caching.

Question 4

Which LLM has the cheapest prompt caching?

Accepted Answer

Depends on volume and prompt size. Claude Haiku and DeepSeek win for high-volume cached workloads; Gemini Flash for very large contexts. Use the ranking table to see your scenario.

Prompt Caching Cost Calculator