How to cut your Z.ai API bill

Mode

Automatic: repeated prefixes are cached with no code change

Automatic cache recognition on mainstream GLM models; no manual configuration.

Cache lifetime

Not specified ("reasonable time limits")

Minimum prompt

Not stated

Cache write cost

No cache-write line item; new content billed at standard prices

Cache read cost

Guide says cache hits are 'usually 50% of standard price', but the pricing table lists cached input at roughly 15-20% of input

Derived from the pricing table. The 15-20% range is computed from the pricing table, Cached Input divided by Input per model (e.g. GLM-5.3 $0.26/$1.4 = 18.6%, GLM-5 $0.2/$1 = 20%, GLM-4.7-FlashX $0.01/$0.07 = 14.3%, GLM-4.5-X $0.45/$2.2 = 20.5%). Only the "usually 50%" guide wording is a quote.

Storage cost

Cached Input Storage listed as 'Limited-time Free' on the pricing table

How to enable

Automatic; check usage.prompt_tokens_details.cached_tokens.

From the docs

“Automatic Cache Recognition: Implicit caching that intelligently identifies repeated context content without manual configuration”

“Cache has reasonable time limits, will recalculate after expiration”

“New content tokens: Billed at standard prices”

“Cache hit tokens: Billed at discounted prices (usually 50% of standard price)”

“| GLM-5.3 | $1.4 | $0.26 | Limited-time Free | $4.4 |”

Sources: docs.z.ai/guides/capabilities/cache, docs.z.ai/guides/overview/pricing

Service tiers on the pages checked: Not stated

Other discounts on the pages checked: Not stated