AllocateWhat is prompt caching?
Prompt caching reuses the computed state of a prompt prefix that repeats across requests, such as a system prompt or tool definitions, so those tokens are not reprocessed and bill at a deep discount. On the current catalog, cached input on GPT-5.5 lists at $0.50 per million tokens against $5 uncached.
Agents benefit most: the system prompt and tool schema are identical on every turn, so an agent with a stable 3,000-token preamble caches it across thousands of requests. GLM 5.2 prices cached input at $0.26 against $1.40 uncached; the pattern is similar wherever a cached rate is published.
Caching changes model rankings on real workloads. A mid-priced model with a published cache rate can beat a nominally cheaper one once your cache hit rate is high; run your own numbers with the hit-rate slider in the price comparison.
Try it: LLM price comparisonRelated terms
Allocate is the cloud inference platform for companies that want to train and run their own models.