Data

Every assessment Hunterson has published, with the shadow period before the first, is free to download and use with credit: "Source: Hunterson Standard Inference, Europe, hunterson.org". Using the figures to settle a contract, value a financial instrument, measure a fund or make up an index needs a free registration, so that everyone relying on them is told directly when the rules change (the licence).

The files

assessments.csv: every day of every live grade, one row a day, with the columns below.
assessments.json: the same, grouped by grade, with the unit, the blend and the licence.
latest.json: the latest assessment only, for programs.
feed.xml: an Atom feed of the latest 30 assessment days, for a feed reader.
The assessments themselves, in the public repository: each day, every provider tested, what it was charged, and why any was excluded.

Units and rounding

Prices are in US dollars per million tokens, to four decimal places, rounded half up by the assessor and never re-rounded. The blended price weights input and output 16 to 1, the shape of the test job. A row whose status is "shadow" was produced before the grade's first assessment and is not an assessment.

What each column means

Column	What it holds
`date`	The assessment day, as YYYY-MM-DD.
`grade`	The grade, such as standard.
`status`	published; not_published, with the reason; or shadow, a day before the grade's first assessment, which is not an assessment.
`reference`	The grade's reference model that day.
`output`	The output price: US dollars per million output tokens. Blank on a day not published.
`input`	The input price: US dollars per million input tokens.
`blended`	Input and output blended 16 to 1, as the test job is: US dollars per million tokens.
`contributors`	Providers that passed every test and contributed to the price.
`panel`	Providers tested.
`reason`	On a day not published, why.
`output_before_discounts`	The output price at the providers' listed prices, before any discount.
`input_before_discounts`	The input price before any discount.
`blended_before_discounts`	The blended price before any discount.
`rollover_ratio`	On the day the reference changes, the ratio for converting a price on the old reference to the new.
`url`	The day's permanent page, the address to cite.
`challenger`	The leading challenger: of the other models tested in the grade, the one with the most providers. For information only.
`challenger_input`	The challenger's input price, assessed in the same way. Not an assessment of the grade.
`challenger_output`	The challenger's output price, assessed in the same way.
`challenger_contributors`	The challenger's contributors.
`challenger_panel`	Providers tested for the challenger.
`challenger_output_ratio`	The challenger's output price divided by the reference's, to four places.

How to cite a figure

Cite the day's permanent page. For example: Hunterson Standard Inference, Europe, 3 November 2026: input $0.2246, output $0.9283 per million tokens. https://hunterson.org/standard/2026-11-03/ Each day's page has this line ready to copy.