Open data

Harpd Product Ranking Data

Name: Harpd Product Ranking Data
Creator: Harpd
License: https://creativecommons.org/licenses/by/4.0/

Harpd publishes structured product ranking data to make product discovery and ranking information easier to explore, analyze, and reference. Every endpoint below is public, machine-readable, and derived from the same first-party data as theHarpd Rank boards.

What is the Harpd ranking dataset?

The Harpd ranking dataset is the public, machine-readable export of Harpd Rank. It currently covers 89 approved products across19 categories, currently led by imgkit. Records contain each product's rank, Rank Points, category, verification status, and public profile URL. The dataset is published under CC BY 4.0 and regenerated from live Rank data on every request.

What these words mean on Harpd

Research, methodology, benchmarks and pricing are different kinds of evidence. On Harpd,research is the evidence and data itself — the datasets on this page.Methodology is how the data is collected and evaluated — the written rules behind every number. A benchmark is a reproducible measurement run under a documented method. Pricing is structured pricing information, verified against official provider pages. We keep them separate so a claim is never confused with its evidence.

The evidence framework

Every claim published on Harpd is supposed to trace through this chain. Where a link is missing, we label it instead of filling the gap.

ClaimExample: “Product X holds rank #N on the overall board with M Rank Points.”
DataThe canonical machine-readable record: /data/rank.json and /data/rank.csv.
MethodologyHow the numbers are produced: Harpd Rank Methodology, also exported as /data/methodology.json.
ResultPublished live on the overall board and the history archive, generated from the same first-party data.
Source / DatasetHarpd Rank first-party data, published under CC BY 4.0 with a versioned field schema.
ReproductionDownload the JSON or CSV, check the schemaVersion, and recompute any board position from the exported records.

The claim → data → method → updated register for every public Harpd claim lives on the Evidence page.

Download the dataset

/data/rank.jsonJSON

Overall ranking: one record per approved product with its current rank, Rank Points, category and verification status.

/data/products.jsonJSON

Product profiles: the ranking fields plus each product’s public website, description and product type.

/data/categories.jsonJSON

Categories: every Harpd Rank category with its slug, canonical page URL and approved product count.

/data/rank-history.jsonJSON

Rank history: every closed monthly and weekly period snapshot with its dates, product count and history page URL.

/data/llm-pricing.jsonJSON

Model pricing: one record per hosted model — list prices per 1M tokens, sourced from official provider pages with source URL and verification date.

/data/llm-pricing.csvCSV

The pricing dataset as a flat CSV for spreadsheets and quick analysis.

/data/benchmarks.jsonJSON

Benchmark data: one record per model × metric (success rate, latency, token use, cost per task / successful task) in long format, with the benchmark’s own last-updated timestamp.

/data/benchmarks.csvCSV

The benchmark dataset as a flat CSV (long format: model, task, metric, score, cost, timestamp).

/data/methodology.jsonJSON

Methodology: the exact rules the ranking numbers follow — Rank Points, boards, entry, disclosure — as machine-readable data.

/data/schema.jsonJSON

Schema: the field dictionary for every endpoint plus what is excluded from the public dataset.

/data/rank.csvCSV

The overall ranking dataset as a flat CSV for spreadsheets and quick analysis.

Last updated: 2026-08-28 (most recent product record change). Pricing data last verified 2026-08-18; benchmark data last updated 2026-08-18.

The benchmark data and its reproducible runner are also mirrored in the open-source repositoryharpd-dev/llm-cost-benchmark(see DATA_DICTIONARY.md andMETHODOLOGY.md there).

Dataset schema

All endpoints share one envelope. Field names are stable;schemaVersion is 1.0. Breaking field changes bump the version rather than silently renaming fields.

Field	Meaning
`schemaVersion`	Version of this dataset schema. Currently "1.0". Field names are stable; breaking changes bump the version.
`generatedAt`	ISO-8601 timestamp of when this response was serialized.
`lastUpdated`	ISO-8601 timestamp of the most recent record-level data change. Moves only when the underlying data changes — never on rebuild alone.
`source`	Where the record comes from: "Harpd Rank" for ranking data, the benchmark id for benchmark records, or the official provider page URL for pricing records.
`methodology`	URL of the Harpd Rank methodology page.
`license`	The dataset is published under CC BY 4.0.
`count`	Number of records in the collection.
`name / slug`	Product display name and its stable URL slug used in profile URLs. `id` equals the slug.
`url`	Canonical Harpd product profile URL.
`website`	The maker’s own public website (products.json only).
`description`	The maker-provided public product description (products.json only).
`category / categoryName`	Harpd Rank category slug and its display name.
`rank`	Current position on the overall board. Products with Rank Points only.
`rankPoints`	Current Harpd Rank Points. One completed Credit spend creates one permanent Rank Point.
`verified`	True when Harpd has verified the product’s listing.
`updatedAt`	Last public update timestamp for the product record.
`provider / model`	Pricing records: billing provider and model display name. Each record also carries `source` (the official pricing page) and `last_updated` (verification date).
`model / task / metric`	Benchmark records: the model under test, the task, and the metric name (success_rate, latency, token counts, cost per task / successful task).
`score / cost / unit`	Benchmark records: the metric value (`score`); `cost` repeats the value in USD for cost metrics and is null otherwise; `unit` states the measurement unit.
`timestamp`	Benchmark records: the benchmark run’s own timestamp, data-derived.
`periodType / periodKey`	History records: "monthly" or "weekly" and the period identifier (YYYY-MM or YYYY-Www).
`startsAt / endsAt`	History records: the closed period’s UTC start and end timestamps.
`productCount`	History records: products recorded in the closed-period snapshot.

How is this data produced?

The dataset is a direct serialization of the Harpd Rank overall board. Products enter the board when they are approved; optional Credits add permanent Rank Points. See Harpd Rank Methodology for how ranks are computed, andproduct categories to browse the boards the dataset mirrors. Harpd Rank statistics are also available on the stats page.

What is not in the dataset?

Only public product and ranking information is published. The dataset never contains user accounts, email addresses, payment or Credit transaction data, moderation notes, or unapproved submissions. Product descriptions are included as provided by their makers.