Portfolio 13 · prefix economics

Prompt cache break-even calculator

Keep cache writes, reads, token-time storage, uncached suffixes, outputs, threshold eligibility, TTL expiry, and unstable-prefix misses visible as independent terms.

Decision this answers: At what reuse level does the stable prefix become cheaper, and is it actually eligible under the selected provider rule and TTL?

Scenario

Decision summary

Uncached baseline

$3780.00exact-derived

All prefix, suffix, and output terms.

[2]

Cached scenario

$2118.70exact-derived

Writes, reads, storage, suffix, and output.

[1][2]

Scenario saving

$1661.30exact-derived

Caching is eligible and cheaper: true.

[1]

Break-even requests

2exact-derived

Suppressed as zero when caching never becomes cheaper.

[1]

Direct-labeled result profile

Uncached baseline Cached scenario Scenario saving Break-even requests

The tables below are the accessible source of truth; bar lengths never carry identity alone.

Cache meter ledger

Cache meter ledger
Candidate	Write	Read	Storage	Eligibility	Operational model	Evidence
Anthropic	Explicit write	Explicit read	TTL-dependent by model	Threshold + stable prefix + TTL	Explicit cache control	published list[1]
Gemini	Model-specific	Cached token rate	Token-time storage	Provider rule	Explicit context cache	published list[2]
OpenAI	Automatic	Cached input rate	Provider-managed	Provider rule	Automatic caching	published list[3]

Official surface versus this site

Anthropic prompt caching

Use the official surface to confirm native prices, counting rules, limits, and product eligibility.

This offline workbench

Provider docs define eligibility, TTL, and cache meters. This page reconciles the complete request family and suppresses a break-even when caching never becomes cheaper.

Sources and methodology

Every result-affecting reference is visible here without JavaScript and is retained in the JSON export.

[1] Anthropic prompt caching

Authority: Anthropic
Native identifier: prompt caching
Unit: tokens and TTL
Evidence: published list
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: Threshold, cache-control, write/read, and TTL rules remain provider-specific.

Open the authoritative source

[2] Gemini API pricing

Authority: Google
Native identifier: gemini-3.6-flash standard caching
Region: global
Unit: USD/1M tokens and token-hour
Evidence: published list
Price kind: list
As of: 2026-08-01
Effective from: 2026-08-01
Retrieved: 2026-08-01
Freshness: fresh
Normalization: The dated input, cached input, output, and storage meters remain independent.

Open the authoritative source

[3] OpenAI prompt caching

Authority: OpenAI
Native identifier: automatic prompt caching
Unit: cached input tokens
Evidence: published list
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: Automatic caching remains operationally distinct from explicit provider cache objects.

Open the authoritative source

Limits and non-claims

Automatic and explicit caching are not presented as operationally identical.
Changed prefixes and expired TTLs are misses, never discounted reads.
Caching economics do not predict latency, quality, or cache availability.