Overview

• Surya is a 366 million‑parameter foundation model trained on nine years of high‑resolution Solar Dynamics Observatory imagery using custom architectures to handle unusually large data.
• Early tests report about a 16% gain in next‑day flare classification and accurate visual predictions of the Sun’s appearance roughly two hours ahead.
• The model, training data and benchmarks are publicly available on Hugging Face, GitHub and IBM’s TerraTorch, with an open SuryaBench dataset for evaluation.
• IBM and NASA say the system can pinpoint likely flare location and intensity, forecast emerging active regions within 24 hours, and estimate solar wind speed and upper‑atmosphere UV effects.
• Experts note that reliable Earth‑impact warnings require validation on the current solar cycle and inputs such as interplanetary magnetic field measurements, with further fine‑tuning and operational integration planned.