Developer API · v1

Historical model ranking data.

Access leaderboard snapshots, model metadata, scores, ranks, tiers, vote counts, confidence intervals, and time series through a public JSON API.

Public JSONNo key requiredDaily snapshots5-minute cache

Historical snapshots

Pass ?at=2026-08-18 to retrieve the leaderboard for a specific date.

Ranking fields

Each record includes rank, score, tier, voter count, confidence bounds, and the complete S–F distribution.

Time series

Request up to ten aligned model series at daily or weekly resolution.

Endpoints

Endpoint reference.

Successful responses contain a data payload and a meta object containing the data window, algorithm version, and generation time.

GET/api/v1/leaderboards

Board directory

Every public category, its leader, coverage, voter count, and canonical URL.

GET/api/v1/leaderboards/{slug}?at=YYYY-MM-DD

Leaderboard at a moment

Reconstruct a complete ranked board on any available date, including confidence intervals and tier distributions.

GET/api/v1/models?provider=&released_after=&q=

Model catalog

Filter releases by provider, release date, or text and discover stable model IDs.

GET/api/v1/models/{id}/history?category=overall,math&interval=day

Full model history

Chart score, rank, tier, voters, confidence, and S–F distribution across one or many boards.

GET/api/v1/timeseries?models=id,id&category=overall

Comparison series

Fetch aligned daily or weekly points for up to ten models in one request.

Example request

Time series response.

Returns aligned daily or weekly points for the requested models and category.

curl '/api/v1/timeseries?models=gpt-5-6-sol,claude-opus-5&category=overall&from=2026-08-11&interval=day'

View JSON ↗

{
  "data": [
    {
      "id": "claude-fable-5",
      "rank": 1,
      "tier": "S",
      "score": 6,
      "voters": 1,
      "confidence": null,
      "distribution": {
        "S": 1,
        "A": 0,
        "B": 0,
        "C": 0,
        "D": 0,
        "F": 0
      }
    },
    {
      "id": "claude-mythos-5",
      "rank": 2,
      "tier": "S",
      "score": 6,
      "voters": 1,
      "confidence": null,
      "distribution": {
        "S": 1,
        "A": 0,
        "B": 0,
        "C": 0,
        "D": 0,
        "F": 0
      }
    }
  ],
  "meta": {
    "apiVersion": "1.0",
    "algorithm": "revision-mean-v1",
    "category": "overall",
    "at": "2026-08-28",
    "source": "persisted ballot revisions"
  }
}

Historical contract

History stays interpretable.

Each point is reconstructed from immutable ballot revisions and identifies the aggregation algorithm that produced it. A methodology change creates a new algorithm version instead of silently rewriting old scores.

Dates are UTC in YYYY-MM-DD form. Unknown IDs return structured errors. Dates without saved ballots honestly return an empty data set.