Open load forecasts for US grid operators, generated by fine-tuning Chronos-2 on historical demand and temperature data.

This project presents a significant disruption to the energy sector's operational forecasting. By outperforming incumbent US grid operators' day-ahead load forecasts, it highlights the potential for advanced machine learning models to optimize critical infrastructure management. The 40% lower Macro MAE translates directly into substantial operational efficiencies, reducing costs associated with over/under-generation, improving grid stability, and enhancing resource allocation. This open-source approach challenges proprietary forecasting models, fostering transparency and potentially accelerating innovation in energy management. The ability to reproduce the benchmark and access the model openly lowers barriers to adoption and validation. This has direct B2B implications for energy traders, utilities, and grid operators seeking more accurate predictive analytics.