Data & API

Free static JSON files, no keys needed, refreshed about every 12 hours.

Endpoints

Plain HTTPS GET; responses are JSON. Relative to this site's address.

GET /data/models.json
Every model: scores, 90% intervals, category scores and ranks, prices, context, modalities, speed.
Download
GET /data/models/{slug}.json
One model's benchmark results and price history. The slug is the last part of a model page URL.
GET /data/benchmarks.json
Benchmarks with category, description, citation, direction (lower or higher is better) and top results.
Download
GET /data/arenas.json
LMArena ratings (CC BY 4.0) for text, webdev, vision, search, documents, images and video, with links to our model slugs.
Download
GET /data/updates.json
Change log: new models, price changes and ranking moves, newest first.
Download
GET /data/meta.json
Sources with licences and retrieval times, counts, category names and the data licence.
Download

Example

curl -s https://<this-site>/data/models.json | jq '.[0] | {name, score, priceIn, priceOut}'

Model fields

slug	Stable identifier used in URLs
name, creator	Model and lab
type	proprietary \| open \| other \| unknown
reasoning	reasoning \| standard \| hybrid \| unknown
released	Release date (YYYY-MM-DD) or null
context	Context window in tokens or null
score, scoreLow, scoreHigh	Overall score with its 90% interval (BenchLM); null when not ranked
rank	Overall rank; null when not ranked
evidence	supported (verified) \| estimated (provisional) \| null
categories, categoryRanks	Score and rank per category key (coding, math, reasoning, agentic, …)
priceIn, priceOut, priceBlended	USD per 1M tokens; blended = (3 × in + out) ÷ 4
openrouterId	OpenRouter model id when the model is available there
inputModalities, outputModalities	e.g. ["text", "image"]
speed	Our measurements: tps, ttftMs, firstAnswerMs, e2e500Ms, n, measuredAt, providers; null when not measured
priceSource	Official (checked on the lab's pricing page) \| BenchLM \| OpenRouter \| null
params, license, downloads, hfId	Open-weights metadata from Hugging Face
arena	LMArena ratings by arena: rating, low, high, rank, votes, variant

Licence and fair use

Compiled data: CC BY-NC 4.0 (includes data from BenchLM.ai). Speed measurements: ours. Non-commercial use only. Credit "ModelBench" and "Data from BenchLM.ai" with links. Commercial use of the BenchLM-derived fields needs a licence from BenchLM.

Prices from OpenRouter are factual list prices; speed measurements are ours. Please cache responses: the files only change when the data refreshes.