WeatherMesh

WeatherMesh is WindBorne Systems' family of AI weather forecast models; the current generation, WeatherMesh-6 (WM-6), was released in June 2026. It targets the gaps in atmospheric observations over oceans, the Southern Hemisphere and the tropics, and the discrete multi-hour update cycles of physics-based numerical weather prediction.

WeatherMesh is a transformer-based model trained on decades of atmospheric data. WindBorne's own AI data assimilation initialises each run from the observations of its Atlas constellation of long-duration Global Sounding Balloons together with satellite data, including geostationary GOES and SEVIRI imagery, the AMSR2 microwave imager and the IASI and CrIS infrared sounders, plus radar composites and surface weather stations. WM-6 Global produces a 128-member ensemble at 0.25° (about 25 km) out to 15 days and refreshes hourly; the WM-6 3 km edition (marketed as HighRes) delivers surface forecasts over the contiguous US every 15 minutes.

Forecasts are delivered through a REST API (gridded Zarr or netCDF and JSON point forecasts), a Python SDK and CLI, webhooks, an MCP server and an interactive insights platform. Derived products include tropical-cyclone tracks, heatwave, coldwave, atmospheric-river and front events, and population-weighted degree days. WindBorne publishes live benchmarks against ECMWF and other AI models, markets WeatherMesh to energy traders, and has received U.S. Navy and Air Force awards to extend the model. A two-week free API trial is available; commercial access is arranged with WindBorne's team.