Free AI model data. No key, no sign-up.

Everything the site publishes in machine-readable form: two leaderboard indexes as JSON and CSV, and the Wire as an RSS feed. Plain GET requests, no API key, and the JSON and CSV files send Access-Control-Allow-Origin: *, so a browser app can read them directly.

The files

ArtificialWatch Bench Index

AI models ranked across the public LLM leaderboards: each model's best-variant score on every board, its percentile within that board, and the composite index (median of percentiles).

/data/bench-index.json · /data/bench-index.csv · the table · rebuilt 2026-10-17

ArtificialWatch Coding Index

AI models ranked across the public coding leaderboards: each model's score on every coding board, its percentile within that board, and the composite coding index.

/data/coding-index.json · /data/coding-index.csv · the table · rebuilt 2026-10-17

The Wire, as RSS

AI model launches, leaks and news, newest first: the latest 50 items, each with a short description and a link to the full item. Paste it into any feed reader, or into the RSS bot in your Slack or Discord.

/wire/feed · the Wire

Try it

The five models at the top of the Bench Index right now:

curl -s https://artificialwatch.com/data/bench-index.json | jq '.models[:5] | map({rank, name, index})'

The files are rebuilt every day and cached for an hour. If a request lands mid-rebuild you get a 503 with {"error":"the table is rebuilding"}; try again a minute later.

What's in the JSON

Both index files have the same shape. The CSV is the same table, one row per model, with a <source>_raw and a <source>_percentile column for each board.

name, description, url: What the file is and the page that shows it as a table.
generatedAt: When the table was last rebuilt (ISO 8601).
license, attribution: The licence URL and a ready-made attribution line.
method: How the index is computed, in one sentence.
sources[]: Each leaderboard the index reads: id, name, url, scale and the board's own asOf date.
models[]: One entry per model, best first: rank, id, name, vendor, the composite index, how many boards rate it, scores (per source id: the raw score and its percentile, or null) and the model's page on ArtificialWatch.

Each model's score on a board becomes a percentile among the models we map on that board; the index is the median of a model's percentiles across the boards that rate it.

Licence and citation

The composite index and the percentile ranks are ours, under CC BY 4.0: use them anywhere, commercially too, with credit. Each board's raw scores belong to that board, and every file names and links its sources so the attribution travels with the data. Cite as:

ArtificialWatch Bench Index, 2026-10-17. ArtificialWatch. https://artificialwatch.com/benchmarks

ArtificialWatch Coding Index, 2026-10-17. ArtificialWatch. https://artificialwatch.com/coding

Model lists, straight from the providers

For the raw lists of models each provider's API serves, and which are new since you last looked, use ai-model-watch, our free, open-source CLI and MCP server. It reads OpenRouter without a key, and OpenAI, Anthropic, Google Gemini, Mistral, xAI and DeepSeek with your own.

Bench Index · Coding Index · New AI model releases · MCP server

sweeping every 60 seconds

Want the launches as they happen? Free.

A free Chrome push the moment a new model answers on a public API, an email about 15 minutes later. Texts or a phone call on paid plans.

Free forever
No card
Unsubscribe in one click