Space weather forecasting is still in its infancy, but there have been enormous advances in the past 15 years. First-principles and physics-based models are now assimilating data, providing forecasts and “nowcasts” of space weather conditions surrounding Earth. Space weather products for end users are growing in sophistication and utility and are often accompanied by useful visual displays (Simpson, 2004). Several sources of space weather forecasts exist today, including research products and operational forecasts from civilian and Department of Defense sources. As the space weather community works toward a comprehensive “Sun-to-mud” forecasting capability, it is natural to expect challenging initial steps, similar to what occurred in terrestrial forecasting more than 50 years ago (Siscoe, 2006). This should not deter the implementation of first-principles-based forecasting for the upper atmosphere, magnetosphere, and ground-induced currents, augmented by data-driven methods where appropriate. Consider, for example, forecasting upper atmosphere disruptions following detection of a coronal mass ejection (CME) near the Sun. The community should seize the moment to establish an ambitious program of forecasting that takes advantage of the lead time available (usually 1-3 days) after a CME has been first detected. Forecasting geomagnetic storms due to high-speed streams is likely less challenging, with potentially longer lead times. The progress made so far has come from transitioning research models into operational environments. As the community works to expand capabilities, it is time to ask whether “transition to operations” is the appropriate paradigm for advancing the field. Once a research model has been transitioned, what are the next steps? Successful implementation of a space weather forecasting capability requires a vibrant research community focused on improving forecasts. Such a community will, of course, help transition models (Tobiska, 2009; Araujo-Pradere, 2009), but it will also continue to develop and improve those models that have already been transitioned. New scientific insights and tools will be developed in the process. In terrestrial weather, the research divisions of the operational centers produce “reanalyses” that are variants of the operational models run in postprocessing to produce long-term consistent estimates of the atmospheric state. These reanalyses have found widespread use in scientific investigations of the atmosphere, demonstrating that activities related to operations can benefit pure research. The operational models all began as research codes, but their use in applied research brings new benefits. Positive feedback from operations to research is not surprising since models are often used to gain scientific insights. Now is a critical time to exercise vision and plan for an era when space weather forecasting undergoes continuous refinement and improvement, providing both operational benefits and new scientific advances. The implementation of a more comprehensive forecasting capability should begin immediately, with the idea of improving forecasts as new scientific knowledge is gained and as operational results are assessed by a broad community. Improving forecasts is an ongoing challenge, requiring enhanced physical understanding as represented in the models, data-driven techniques where needed, and effective observational systems. Advancements in forecasting require both the dedicated involvement of the scientific community and the existence of an applied research community. To achieve this, a stable but evolving forecasting infrastructure, accessible to the broader scientific community, is required. Such an infrastructure can serve as a focal point for model development and expanded observational systems. The space science community should give serious thought to how we will transition from an era of “first implementation” to an era of continuous improvement. So much has been accomplished. Let us start preparing today for the next great frontier in space weather research. The author wishes to acknowledge Attila Komjathy for a careful reading of this opinion. This research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with NASA. Anthony J. Mannucci is supervisor of the Ionospheric and Atmospheric Remote Sensing Group at Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., USA. E-mail: tony.mannucci@jpl.nasa.gov.