API v1Updated 2026-10-16

Rankings API

GET /api/rankings — public token usage per model and author, trending and fastest models, the data behind the LLM rankings page.

GET /api/rankings is the public, unauthenticated feed behind the LLM rankings page. It reports how many tokens each model served on zrouter over a rolling window, the market share of each model author, trending models, and the fastest models by median output speed. Responses are aggregates only: no prices, request or user counts, or provider details.

Endpoint#

GET/api/rankings

Auth: public

Token usage rankings for one rolling window. Cached for 60 seconds (Cache-Control: public, max-age=60).

Parameters

windowstringqueryoptionaldefault: week

Rolling window ending now: last 24 hours (hourly buckets), 7 days or 30 days (daily buckets), or everything since the first recorded day (weekly buckets, Monday start).

Allowed values:dayweekmonthall

modalitystringqueryoptionaldefault: text

Only text models are ranked today. Image models are not included.

Allowed values:text

Request Example

curl https://api.zrouter.dev/api/rankings?window=week

Responses

200Rankings for the requested window (abridged).

{
  "schema_version": 1,
  "generated_at": "2026-10-16T08:00:00Z",
  "window": "week",
  "modality": "text",
  "range": {
    "from": "2026-10-09T08:00:00Z",
    "to": "2026-10-16T08:00:00Z",
    "previous_from": "2026-10-02T08:00:00Z",
    "data_since": "2026-09-03"
  },
  "totals": { "tokens": 171534300000, "previous_tokens": 155000001000 },
  "leaderboard": [
    {
      "rank": 1,
      "public_name": "deepseek-v4.1-flash",
      "display_name": "DeepSeek V4.1 Flash",
      "author": "deepseek",
      "listed": true,
      "tokens": 111300000000,
      "previous_tokens": 101000000000,
      "change_pct": 10,
      "is_new": false
    }
  ],
  "other": { "tokens": 315000000 },
  "series": {
    "bucket": "day",
    "keys": ["deepseek-v4.1-flash"],
    "points": [
      { "start": "2026-10-08T17:00:00Z", "partial": true, "values": { "deepseek-v4.1-flash": 7950000000 }, "other": 22500000 }
    ]
  },
  "authors": [{ "author": "deepseek", "tokens": 111300000000, "share_bp": 6489 }],
  "trending": [
    { "public_name": "deepseek-v4.1-flash", "display_name": "DeepSeek V4.1 Flash", "author": "deepseek", "tokens": 111300000000, "change_pct": 10, "is_new": false }
  ],
  "fastest": [
    { "public_name": "deepseek-v4.1-flash", "display_name": "DeepSeek V4.1 Flash", "author": "deepseek", "tps_p50": 91.2, "ttft_p50_ms": 2365, "samples_bucket": "10k+" }
  ]
}

400`invalid_request`: unknown `window` or `modality` value.

404`not_found`: rankings are not enabled on this deployment.

500`internal_error`: usage data could not be read; retry later.

Fields#

Field	Meaning
`range.from` / `range.to`	Window bounds (UTC, minute precision). `to` is when the snapshot was taken.
`range.previous_from`	Start of the comparison period of equal length right before `from`; `null` for `all`.
`range.data_since`	First day with recorded usage (Asia/Jakarta calendar date).
`totals.tokens`	All text tokens in the window: the leaderboard plus `other.tokens`.
`leaderboard[].listed`	Whether the model is currently in the public catalog. Retired models keep their history but are not listed.
`leaderboard[].change_pct`	Integer percentage vs the previous period; `null` when `is_new` is true or for `all`.
`other.tokens`	Tokens of models that are not shown by name.
`series`	Chart data: per bucket, tokens of the top models (`keys`) and the rest in `other`. `partial` marks a bucket cut by the window edge.
`authors[].share_bp`	Market share in basis points (1/100 of a percent). Always sums to exactly 10000.
`trending`	Visible models with at least 1M tokens in the last 7 days that are new or growing vs the 7 days before; up to 10.
`fastest[].tps_p50`	Median output tokens per second of successful streamed requests over 7 days, end to end (includes time to first token).
`fastest[].samples_bucket`	Coarse sample size: `200+`, `1k+` or `10k+`. Models need at least 200 streamed requests to appear.

How tokens are counted#

Tokens = input + output of every inference request on zrouter, including failed requests that consumed tokens. Cache reads are not counted. Health-check traffic is excluded.
Days follow Asia/Jakarta (UTC+7). Daily and weekly buckets start at 00:00 WIB.
Totals match the status page. The all window equals the lifetime tokens on the status page, minus image models.
Fresh for one minute. Results are computed at most once per minute per window and served from cache.