The Kingy.ai AI API Price Index is a source-audited dataset of publicly documented AI API prices. It is designed to answer a narrower, more reproducible question than a conventional pricing roundup: what price is officially documented for this exact callable offering, through this seller and route, in this region or documented non-regional scope, under this service tier, billing mode and context band?
Current methodology version: 1.3.0. Canonical release: kapi-price-index-2026-08-23T042959Z.
Unit of observation
An offering is not a model family name. Its identity includes:
- access provider or seller;
- exact API model ID;
- first-party or managed-cloud route;
- endpoint;
- region, or explicit documented non-regional routing semantics;
- service tier and billing mode;
- modality and price component;
- context minimum and maximum, when tiered;
- effective start and end, when officially stated.
Aliases, promotional rates, cloud-hosted copies and first-party versions are never merged merely because their marketing names look alike.
Evidence standard
Accepted records require current official public documentation supporting every material identity and price field. Each source has a registry record and an immutable capture record containing the observed URL, status, capture timestamp, response headers, normalized text locator and SHA-256 content hash.
Search snippets, third-party roundups and recollection can help locate a source. They cannot support an accepted price.
What the index does not infer
Observation time is not treated as an effective date. A global-looking page is not treated as proof of global availability. A model family label is not treated as an API model ID. A “batch available” statement is not treated as proof that every model has a batch price.
Missing regions, ambiguous aliases, incomplete price tables, future rates and material disagreements go to the public hold ledger. Holds are excluded from accepted comparisons until official evidence resolves them.
Normalization
Native token prices are normalized to US dollars per one million tokens. The standard representative workload is one million input tokens plus 250,000 output tokens:
workload_cost = input_price + (0.25 × output_price)
The projection uses matching input and output components from one exact price identity. It does not mix context bands, routes or tiers. Cache reads, cache writes, batch, flex, priority and other service modes remain separate records.
Freshness and change control
Official sources are eligible for daily acquisition and diffing. Queued changes receive weekly editorial review. Explicit vendor announcements target review within one business day. A versioned snapshot is expected at least monthly while the index is active.
Automation can acquire, hash, extract, normalize, diff and validate. It cannot authorize publication. A human resolves conflicts and approves each release.
Validation
Each release must pass Draft 2020-12 schema validation plus referential, arithmetic, duplicate-ID, endpoint, route, region, tier, artifact-hash and source-hash checks. The public page is also checked for links, structured data, accessibility and responsive behavior.
Licensing and citation
The dataset and generated charts are licensed under CC BY 4.0. Kingy.ai editorial text remains separately copyrighted.
Suggested citation: Kingy.ai, “AI API Price Index 2026,” release kapi-price-index-2026-08-23T042959Z, methodology 1.3.0, https://kingy.ai/kapi/price-index/.
Download the canonical release and integrity files, or inspect the release changelog.