RoninForge/ai-price-index

AI Price Index

An open, dated, first-party-sourced record of AI model API prices over time.

Provider pricing pages change quietly with no changelog. This dataset is the changelog. Every price carries the date it became valid and the date it was last verified against the official source, so you can price historical token usage point-in-time instead of extrapolating from today's rate.

• 356 price records, 108 models, 11 providers: Anthropic, OpenAI, Google, Mistral, xAI, DeepSeek, Cohere, Amazon Nova, AI21, Alibaba Qwen
• A first-party source URL on every price
• Maintained via a daily automated check against official pricing pages; every change is human-reviewed before it lands
• License: CC BY 4.0. DOI: 10.5281/zenodo.20730240

Load it

from datasets import load_dataset

ds = load_dataset("RoninForge/ai-price-index", split="train")
print(ds[0])

Schema

One row is one price observation (a model's input or output rate during a dated validity window).

column	description
provider	Provider slug (e.g. anthropic, openai).
model_id	Provider model id (e.g. claude-opus-4-5-20251101).
variation	input or output tokens.
unit	Price unit. Currently usd_per_mtok (USD per 1M tokens).
price_usd	The price, in USD, for that unit.
effective_from	Date this price became valid (ISO YYYY-MM-DD).
effective_to	Date this price stopped being valid, or empty if it is the current price.
last_validated_at	Date the price was last checked against the official source.
source_kind	How it was sourced (e.g. provider_live).
confidence	verified (first-party-confirmed) or inferred (price known, exact date estimated).
aliases	Other ids the model is known by (semicolon-separated).
source_url	First-party source for the price.
notes	Free-text context.

To get the current price of a model, filter rows where effective_to is empty. To price usage on a given date d, take the row where effective_from <= d and (effective_to is empty or effective_to > d).

Files

• ai_price_index.csv: the full flat table (loaded above).
• ai_price_index.json: the same records as JSON.

The methodology, per-model price history, and changelog live on the dataset home page. The bitemporal ledger and the tooling that builds it live in the GitHub repository.

Attribution

CC BY 4.0 means you can use this data anywhere, including commercially, for free. Credit it like this:

AI Price Index by RoninForge (https://roninforge.org/data/ai-price-index/), CC BY 4.0. Values validated .

Keep the validation date in. Prices change and this dataset supersedes values without deleting history, so a dated citation stays checkable years later.

Maintenance and methodology

A bot diffs each provider's official pricing page daily and flags drift. No price changes automatically: a human verifies every change against the first-party source before it is merged, because pricing is exactly the kind of data you do not want a scraper silently guessing on. Corrections and new-provider requests are welcome via GitHub issues and pull requests.

Links

• Dataset home, methodology and history: https://roninforge.org/data/ai-price-index/
• Citable DOI (concept, always latest): https://doi.org/10.5281/zenodo.20730240
• Build system, issues, PRs: https://github.com/RoninForge/ai-price-index

Citation

@misc{roninforge_ai_price_index,
  author       = {{RoninForge}},
  title        = {{AI Price Index: dated, first-party AI model API prices over time}},
  year         = {2026},
  doi          = {10.5281/zenodo.20730240},
  howpublished = {\url{https://roninforge.org/data/ai-price-index/}},
  note         = {Data licensed CC BY 4.0.}
}