Kingy AI
Subscribe, 2.1M subscribersWork with Kingy
AI News

AI Model Price Stability: A 31,946-Observation Audit

Kingy.ai Research · Dataset version 2026-08-16

Kingy retained 31,946 price observations from 1,861 unique archive captures across 152 raw series. After withholding 18 scope-ambiguous series, 134 series support 157 price states and 23 machine-detected value transitions.

Read this distinction first. The 23 values below are machine-detected transitions, not 23 automatically verified price changes. Kingy’s editorial price ledger applies additional model-identity, release-state and date-bracket gates and currently publishes 10 verified episodes.
31,946price observations
1,861unique archive captures
152raw series
157scope-safe price states
23machine transitions

Place this audit in the wider decision context with Kingy’s Model Economics hub, covering workload costs, pricing evidence and migration planning.

Observation quality boundary

18 scope-ambiguous series are excluded from public series rows.

Machine transitions by observation month

Month is the first retained observation of a different normalized value, not necessarily the date a vendor changed its page.

Provider coverage

Provider profile Models Eligible series Observations States Transitions Official source
Claude API 17 48 15,826 58 10 Official pricing
Gemini API 19 35 7,461 45 10 Official pricing
OpenAI API 18 51 1,759 54 3 Official pricing

Methodology and refusal rules

A machine-detected transition is counted only when consecutive scope-unambiguous observations in the same series have different normalized values. It is not a verified editorial price-change claim.

Any series with multiple values at one timestamp is withheld because the retained contract lacks the provider scope dimension.

  • An observation is evidence of a captured published price, not a guarantee of universal availability or eligibility.
  • Regional, plan, context-window, batch, caching and other scope may be absent from the retained contract.
  • Held scope-ambiguous series are excluded from transition counts and publication candidates.
  • The archive does not claim complete coverage of every provider, model, component or date.

The editorial methodology documents the stricter rules used before Kingy calls a candidate a published price-change episode.

Download and cite

Versioned JSONDerived price-state CSVRaw observation CSV

Kingy.ai, “AI Model Price Stability: A 31,946-Observation Audit,” dataset version 2026-08-16.

Published price-history series

The defensible subset of the 15 changed-series candidates is now available as source-linked, component-level history. Three identity or date-bracket failures remain unpublished.

Browse all Kingy AI research

Release v2026.08.1 · Published · Coverage 2024-06-20/2026-08-02 · Cadence Living dataset with immutable versioned snapshots

A longitudinal research package documenting a 31,946-observation AI model pricing audit, the conservative scope gate applied to archive captures, a normalized public dataset of price states, and machine-detected transition candidates.

Research lead: Curtis Pyke (Kingy AI). Independence: Independently funded. No external funding or sponsorship; the work was conducted by Kingy AI. Curtis Pyke declared no relevant financial or personal conflicts for these research products.

Methodology

Extracts archive observations, applies a scope-conflict gate, reduces eligible observations to price states, and flags transitions for editorial verification.

Population: 31,946 retained observations from 1,861 archive captures across 152 raw vendor/model/component series.

Uncertainty: Capture timing bounds observation rather than exact commercial effective dates, and missing or ambiguous pages may conceal intermediate states.

Limitations

  • Archive coverage is incomplete
  • Detected transitions are not automatically verified changes
  • Vendor scope and tiers may be ambiguous
Horizontal bars show anthropic with 15,826 eligible observations, google with 7,461 eligible observations, openai with 1,759 eligible observations.
Eligible observations by provider. CSV data · SVG chart · PNG chart

Versioned downloads

  • model-price-stability.csv data · SHA-256 97badbe5cf882d068e6efabff72f4801c00fa414de218c9a8632cb6b4a6c522e
  • model-price-stability.json data · SHA-256 d57226e39da51e85a8e5eeb0e3d3389ab34185d055b17b8273977e4fe7f3cb92
  • research-report.json report · SHA-256 7d55c88efc68c30a46b0d457ec6ce12249d40ff5f726b1206f288dec1f6074dc
  • methodology.md methodology · SHA-256 fa60c419d562f990dabf74045ad2520b1b48610aa0032fe5f603a896bbd9fed8
  • data-dictionary.csv schema · SHA-256 7ceb2afcc44ae1d8ebbe1f9afc21365be9d541f5e15ab36a5e6b96c7b9649cb3
  • sources.csv source-manifest · SHA-256 66ed32cec9784f4c7c3c76a7acdafd4936f878ce532fdb1e8f9cf3e9876cc822

Cite this research

Cite the exact files used with 10.5281/zenodo.22294510. Use the concept DOI 10.5281/zenodo.22288610 when referring to the evolving product across versions.

Research roles

  • Conceptualization
  • Data curation
  • Formal analysis
  • Methodology
  • Validation
  • Visualization
  • Writing – original draft

Disclosure and corrections

Curtis Pyke and Kingy AI controlled the questions, methods, analysis, conclusions, wording, and publication decision. No cash or in-kind consideration was received from the vendors or products covered by these research products. No covered vendor or other external interested party had prepublication review or veto rights.

AI systems assisted with research-product packaging, code generation, and validation. Curtis Pyke reviewed the release decisions and accepts accountability for the published work.

No independent external review was performed. Independence does not eliminate source incompleteness, archive gaps, interpretation error, or the documented methodological limitations.

Corrections will be documented in versioned errata or a new immutable release; released files will not be silently replaced.

Licenses: data CC BY 4.0; code MIT; content and charts CC BY 4.0.