30 forecast models. One clean API.

Not enabled yet

isn't part of an organization with access to the Forecast API. Ask us to enable your company's email domain — every colleague on it gets access at once.

Contact info@benchmarklabs.com to activate your team.

Set a new password

Your API key

Your locations

The first 10 unique coordinates you pull are locked to your account automatically (rounded to 2 decimal places ≈ 1 km — nearby jitter reuses the same slot). Pull these as often as you like. Locations are shared account-wide when access is enabled for your whole domain.

Your usage

Quick start

One endpoint. Pass a latitude, a longitude, a model, and your key:

Your key can go in the X-API-Key header, an Authorization: Bearer header, or the ?key= query parameter — whichever is easiest in your tooling.

Endpoint reference

GET /v1/forecast

Models

Every model returns one value per hour, in UTC, for its full horizon (a model's native step can be coarser at long lead times — values are served as a uniform hourly series). GET /v1/models returns this list as JSON (no auth needed).

Response

{
  "location": { "lat": 39.74, "lon": -104.99 },
  "model": "benchmark",
  "units": "metric",
  "generated_at": "2026-08-04T22:00:00Z",
  "offsetMinutes": -360,            // location's UTC offset
  "hourly": [
    {
      "time": "2026-08-04T23:00:00.000Z",   // UTC
      "tempC": 23.0,          // °C   (imperial: tempF, °F)
      "rh": 38,               // relative humidity %
      "windMs": 3.1,          // m/s  (imperial: windMph)
      "windDir": 210,         // degrees, meteorological
      "precipMm": 0,          // mm/hour (imperial: precipIn)
      "snowMm": 0,            // mm/hour (imperial: snowIn)
      "precipChance": 10,     // %
      "uvIndex": 6,
      "feelsLikeC": 23.5      // (imperial: feelsLikeF)
    }
  ]
}

Fields a model does not provide come back as null (e.g. NWS has no hourly precipitation amounts or UV).

Time zones

Every time value is UTC (ISO-8601, Z suffix), top-of-hour, for every model and observation source — join series from different models directly on time. Forecast series start at the current hour: no past values, ever. offsetMinutes is the location's current UTC offset — local time = time + offsetMinutes minutes. Station observation responses also include the IANA timeZone name (e.g. America/Los_Angeles). One caveat: offsetMinutes is the offset right now — if a daylight-saving change falls inside a 15-day horizon, convert with a proper tz library (Python zoneinfo) using the timeZone name instead of adding a fixed offset.

Errors

Limits

Trial default: 10 locations, unlimited pulls on them up to 10,000 calls/day, soft limit 60 calls/minute. Forecasts update continuously; pulling more than once per 10 minutes per location returns the same data. Need more locations or volume? Talk to us.

Observations — actual measured conditions

Forecasts across models tell you the spread (how much the models disagree — a direct uncertainty signal). Observations tell you the truth: what actually happened, so you can score which model has been most accurate at your locations lately.

GET /v1/observations

Station response

{
  "location": { "lat": 32.72, "lon": -117.16 },
  "source": "station",
  "station": {
    "id": "KSAN", "name": "San Diego International Airport",
    "lat": 32.73, "lon": -117.18, "distance_km": 2.1
  },
  "timeZone": "America/Los_Angeles",
  "offsetMinutes": -420,
  "hours": 48,
  "hourly": [
    { "time": "...", "tempC": 22.8, "rh": 64, "windMs": 4.1, "windDir": 290,
      "precipMm": 0, "dewPointC": 15.6, "pressureHpa": 1013.2 }
  ]
}

Score the models against reality

Pull yesterday's forecasts each day, store them, then compare against station actuals — after a couple of weeks you know exactly which model to trust at each of your locations:

Code examples

Python

Excel (Power Query)

Data → Get Data → From Other Sources → Blank Query, then:

Compare all models at one location