The Open-Meteo API has traditionally provided weather data as a continuously updated seamless time-series. This approach is ideal for end-user applications but fundamentally misses the individual run structure – the full forecast horizon of a model at a specific initialisation time. The full forecast horizon of past weather models is useful for research post-processing AI/ML and backtesting workflows.
The Single Runs API fills that gap.
What it provides
The API exposes the complete forecast horizon of any individual model run from the past years, selected by initialisation time using the “run=” parameter:
GET https://single-runs-api.open-meteo.com/v1/forecast
?latitude=52.52&longitude=13.41
&run=2024-06-01T00:00
&hourly=wind_speed_10m,wind_speed_100m,wind_speed_200m
&models=ecmwf_ifsThe parameter set is otherwise identical to the Weather Forecast API — all the same variables, models, units, pressure levels, and output formats are supported. The only addition is the required “run=” parameter.
The weather forecast for the requested coordinates at the specified initialisation time is returned. The “models=ecmwf_ifs” parameter allows you to limit the forecast to a specific weather model.
Data for more than one location: Bounding Boxes
The Single Runs API isn’t restricted to a single location; it allows users to query smaller geographical areas. By using the parameter “&bounding_box=44,7,45,8” instead of latitude and longitude, users can retrieve all grid cells within a chosen weather model.
Bounding boxes can return up to 1000 grid cells and are only applicable to individual domains. Models like “best_match” or “icon_seamless” combine multiple weather models at various resolutions and therefore aren’t compatible with bounding boxes.
Model Initialisation Time
The weather model initialisation time is the UTC reference time when observations are taken, not when the forecast output becomes publicly available. After initialisation, the model needs additional computation time before results are distributed. This typically takes 4–6 hours for global models like the ECMWF IFS and GFS and 1–3 hours for regional models. Therefore a run initialised at 00 UTC is generally accessible from around 04–06 UTC onwards.
Setting the model initialisation time to “2026-05-15T00:00Z” doesn’t mean the forecast was available at 0:00 UTC in the API. It simply indicates when the model observations were taken. Weather models initialise with 0z measurements and still require a couple of hours to compute before their forecasts are available.
Archive coverage
The ECMWF IFS HRES dataset at native 9 km resolution (O1280 grid) is the main focus. Individual runs at 00, 06, 12 and 18 UTC are archived from March 14, 2024. This provides over two years of retrievable runs with 15-day forecast horizons. Starting May 12, 2026 at 06 UTC, runs will use the updated IFS Cycle 50R1.
For all other models — including GFS, ICON, Météo-France AROME, and the remaining Forecast API suite — archival begins April 2026.
IFS HRES 9 km (O1280 grid): March 14, 2024 onwards
Other Weather Models: April 2026
Over the coming months, we’ll be integrating additional historical data from other weather models such as GFS and HRRR.
While more data is desirable, weather models are constantly upgraded and increasingly accurate measurements become available. This, however, inadvertently alters their behaviour and introduces minor inconsistencies. AI and ML workflows trained on inconsistent data can introduce unwanted artefacts during the interference step. In the case of ECMWF IFS, we deliberately integrated only historical weather data from March 14 onwards and used model versions Cycle 49R1 and 50R1.
How this differs from other historical weather APIs
Open-Meteo has been providing historical weather already through different APIs. Currently, all APIs form continuous time-series and don’t allow access to individual weather model runs.
Historical Forecast API (data from 2021): Seamless hourly time-series built by stitching each run’s first few hours. The Historical Forecast API is the perfect choice for obtaining a single high-quality time series of past conditions close to measurements. It’s ideal for analysing gridding bias in weather models.
Single Runs API (Sep 2025, ECMWF March 2024): The complete forecast horizon of one specific run, unmodified. The Single Runs API is ideal when the identity of a run is crucial, particularly for AI and ML model training.
Previous Runs API (from Jan 2024): Variables at a fixed lead-time offset (1–7 days ahead) across all runs. The Previous Model Runs API falls somewhere between the Historical Forecast API and the Single Runs API. It’s useful for evaluating the lead-time error of weather models over years of data within seconds but lacks the fine-grained control of the Single Runs API.
Operational Processing - Low Latency
Forecasts from weather models are processed as soon as data has been published by weather services. Open-Meteo introduces a small delay because we also need to download process and distribute data across API servers in Europe and North America.
For ECMWF IFS, we retrieve data via the “pre-schedule delivery” to minimise latency. We redistribute it through an AWS open-data sponsorship, achieving a 1-minute delay for each forecast time-step. Data via the Single Runs API becomes available shortly after the final weather model time-step is published.
Getting started
The Single Runs API is available at https://single-runs-api.open-meteo.com/v1/forecast. A full interactive parameter builder is on the Single Runs API documentation page.
The free tier applies the same rate limits as the main weather forecast API; no API key is required for non-commercial use. For commercial use or more API calls, the “API Professional” subscription starting at 99 Euro per month is required. More information is available on our pricing page.
Features like the Single Runs API are developed in close collaboration with our “Enterprise API” customers, who benefit from dedicated support, SLAs, higher API call volumes, and early access to new and experimental features. For professional data scientists and machine learning engineers, getting easy programmatic access to individual weather model runs has historically been a significant challenge — driven by the sheer volume of data involved and the complexity of working with raw model output formats. Through this ongoing work with our enterprise customers, we are excited to now bring the Single Runs API to a wider audience and look forward to hearing your feedback.
If your company is also interested in priority support, custom SLAs, or early access to upcoming features, we would love to hear from you. Get in touch at info@open-meteo.com and our team will be happy to discuss how the Open-Meteo API can be tailored to meet your needs.
very interesting and useful API!
Hi team, once again, thank you for all your work. I had a quetion about this framework of the single run. I acknowledge that it might be difficult to extend this logic to ensemble models (i.e individual members). However, would it be possible to extend it to the mean of ensemble (average of 2m temp across members, same with 100m wind speed etc) ? The result would be the time serie of the average of members but accessible with an as_of_date and a run_id.
Thanks again, best