Claude Certified Architect – Foundations Practice Question 60: Cache Below Minimum Length
Claude Certified Architect, Foundations · Test 6 · question 10 of 10
A team calls Claude Haiku 4.5 with a system prompt and tool definitions totaling about 2,800 tokens, and places cache_control on the last system block. Every call succeeds, but the usage object always shows cache_creation_input_tokens of 0 and cache_read_input_tokens of 0, and the cost never drops. Requests are sent several minutes apart from a single worker. What is the most likely cause and fix?
Show answer and explanation
Answer:
C is correct. Each model has a minimum cacheable prompt length, and the docs list 4,096 tokens for Claude Haiku 4.5. Shorter prompts are processed without caching and no error is returned, even with cache_control set. If both cache_creation_input_tokens and cache_read_input_tokens are 0, the prompt was not cached, likely for this reason. The fix is to expand the cached prefix past the threshold (often worthwhile because reads cost far less) or to use a model with a lower minimum.
A is wrong because an expired entry would still show a nonzero cache_creation_input_tokens when it was rewritten; zeros on every call mean nothing was ever cached. B is wrong because a breakpoint on a system block is valid; breakpoints mark the end of a cacheable prefix wherever it ends. D is wrong because a race between concurrent requests delays availability of an entry, but a single sequential worker is not racing, and the first write would still be counted in cache_creation_input_tokens.
Reference: Prompt caching