Data API
Live AI model benchmark data as JSON: rankings, the global intelligence index, degradation alerts, drift signatures and provider reliability.
curl "https://aistupidlevel.info/api/v1/models?period=24h" \ -H "X-API-Key: $ASL_DATA_KEY"
{
"success": true,
"period": "24h",
"sort_by": "combined",
"data": [
{ "name": "example-model", "currentScore": 80.4,
"confidenceLower": 76.2, "confidenceUpper": 84.6, … }
]
}01
These endpoints used to be open, and plenty of people built genuinely good things on them — status bars, dashboards, monitoring bots. That open access also got used to re-publish the rankings wholesale as other people's leaderboards, with no attribution and no way for us to tell one from the other.
A key fixes exactly that: it costs nothing, takes about thirty seconds, and turns an anonymous user-agent string into something we can rate-limit, contact and — if it turns out to be a mirror rather than a client — revoke. If you were using the old open endpoints, create a key and add one header. Nothing else changes.
02
Send your key in the Authorization header. An X-API-Key header works too if a bearer token is awkward in your client.
curl -H "Authorization: Bearer asl_live_your_key_here" \ "https://aistupidlevel.info/api/v1/models?period=24h&sortBy=combined"
https://aistupidlevel.info/api/v1
The same key works in the n8n node (n8n-nodes-aistupidlevel, Benchmark operations) and the MCP server (@aistupidlevel/mcp) for Claude, Cursor and VS Code. Setup for both is in the router documentation.
03
The Data API covers the models we benchmark through their makers' own APIs. Models hosted by independent inference providers are not part of it.
/api/v1/modelsCurrent rankings with scores, trends and confidence intervals.
periodsortBy/api/v1/models/:idDetail for one model, including per-axis breakdown.
none/api/v1/models/:id/historyScore time-series for one model, on the board you name.
periodsortBy/api/v1/indexGlobal intelligence index — the aggregate score across all tracked models, with history.
none/api/v1/alertsActive performance alerts for models scoring below their baseline.
none/api/v1/incidentsRecorded degradation incidents.
periodlimit/api/v1/providersProvider-level availability and status.
none/api/v1/analytics/degradationsModels currently degrading, with magnitude and direction.
period/api/v1/analytics/recommendationsWhich model to use for each task type, over the period you ask for.
period/api/v1/analytics/provider-reliabilityPer provider: measured uptime, live status, recorded incidents and a trust score (uptime minus 5 points per incident).
period/api/v1/analytics/transparencyFull scoring transparency data.
period/api/v1/drift/:idDrift signature for one model: Page-Hinkley statistic, regime and per-axis trends.
none/api/v1/meYour key, its tier, and your current quota.
none04
periodlatest, 24h, 7d, 1m (30d is accepted for 1m). The latest run is part of the paid plans: a paid key defaults to latest, while a free key defaults to 24h and gets a 403 with latest_requires_plan when it asks for latest. On a free key, /models/:id leaves out latestScore.sortBycombined, reasoning, speed, price, coding (also called 7axis), tooling (tool use). Defaults to combined. History takes combined, coding, reasoning or tooling.Invalid valuesperiod or sortBy returns 400 with invalid_period or invalid_sort and the accepted values. It is never swapped for a default.05
Every successful response uses the same envelope.
{
"success": true,
"version": "1.0.0",
"generated_at": "2026-09-04T07:48:00.000Z",
"license": "Attribution required — cite aistupidlevel.info as the source.",
"period": "24h",
"sort_by": "combined",
"count": 24,
"data": [
{
"id": "123",
"name": "example-model",
"provider": "openai",
"currentScore": 80.4,
"trend": "up",
"status": "good",
"confidenceLower": 76.2,
"confidenceUpper": 84.6,
"lastUpdated": "2026-09-04T06:54:01.041Z"
}
]
}06
| Tier | Per day | Per minute | How to get it |
|---|---|---|---|
| Free | 10 | 1 | Sign up and create a key |
| Pro | 10,000 | 60 | Included with Pro |
| Developer | 25,000 | 120 | Included with Developer |
| Teams | 100,000 | 300 | Included with Teams |
| Enterprise | 250,000 | 1,000 | By arrangement — see below |
Free is an evaluation tier. Ten calls a day is enough to wire a client up and see real data come back — it is not enough to poll on a schedule. Anything that refreshes by itself needs Pro.
Every response carries X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset. The daily quota resets at 00:00 UTC. Over quota returns 429 with a JSON body explaining which limit you hit.
Only answers count against the daily quota. A refusal or an error (an invalid parameter, a paid-only view on a free key, an unknown model, or a 429 for being over quota) is not counted.
Coding scores change at most every four hours and reasoning and tool-use scores once a day, so cache on your side rather than polling. A request every few hours keeps you current, and it keeps a Pro key nowhere near its ceiling.
07
Enterprise is not sold through a checkout — there is nothing to buy on this page. If you need volume beyond Pro, a commercial redistribution licence, or a guarantee we will not change something under you, get in touch and we will sort out limits that fit what you are actually building.
08
| Status | Code | Meaning |
|---|---|---|
| 400 | invalid_model_id | The model id in the path is malformed. |
| 400 | invalid_period | period is not one of the accepted values. The body lists them. |
| 400 | invalid_sort | sortBy is not one of the accepted values (or not one history takes). The body lists them. |
| 401 | api_key_required | No key was sent. |
| 401 | invalid_api_key | The key is unknown or has been revoked. |
| 403 | latest_requires_plan | A free key asked for period=latest. Use 24h, 7d or 1m, or a paid plan. |
| 404 | model_not_found | No such model. |
| 429 | quota_exceeded | Daily quota spent. Resets at 00:00 UTC. |
| 502 | Bad Gateway | Our side failed to produce the data. Not caused by your request and not counted against your quota. Retry later; if it persists, contact us with the time and path. |
09
Attribution is required. Anywhere you display this data, cite aistupidlevel.info as the source with a link.
Do not republish the rankings as your own leaderboard, and do not resell the data. Use one key per application rather than sharing a key across products. Keys that behave like a mirror rather than a client can be revoked without notice.
Building something that needs to run on a schedule? That is what Pro is for. Building something bigger than Pro, or that redistributes the data? Talk to us first — we would much rather agree terms with you than find out from the logs.
It costs nothing and takes about thirty seconds. Then add one header.