Abstract:This paper uses the BCC model to estimate and evaluate the efficiency of the patent output of enterprises, universities and research institutions in Beijing according to the data from 2005 to 2016.Then a dummy variable model, including the three Innovation Subjects, is constructed to analyze the robustness of the difference between their patent output efficiency. After that, the efficiency of the three is analyzed by the DEA-malquist index method to explore details and reasons of dynamic changes of their respective efficiency. Finally, by using Grey relational model, the impact of their respective internal investment structure on the efficiency is explored. The result shows that the comprehensive efficiency of patent output of enterprises is the highest one, followed by universities and research institutions and the overall efficiency of enterprises, universities and research institutions are increasing year by year. But the main engine of enterprise R&D efficiency is technical progress, while the improvement of universities and research institutions efficiency is caused by technical progress and reduced scale efficiency. In addition, there is a big difference between the effect of the input structure and resource types of the three on their patent output efficiency. This paper puts forward some policy Suggestions from the perspective of dynamic optimization.