Cursor, a U.S.-based AI coding company valued at $29.3 billion, quietly launched its latest model, Composer 2, on top of an open-source project from Chinese firm Moonshot AI. This revelation came after an online user pointed out that the model’s underlying code identified Moonshot’s Kimi 2.5 as its foundation.
The Disclosure
Cursor’s VP of developer education, Lee Robinson, later confirmed the origin, stating that while the model started with Kimi 2.5, approximately 75% of the training was done independently using proprietary compute resources. He claims this resulted in performance significantly different from the original Kimi model.
The Partnership
Moonshot AI itself confirmed the arrangement, stating Cursor used Kimi “as part of an authorized commercial partnership” through Fireworks AI. The company framed it as a success story for the open-source AI ecosystem, celebrating the integration of its model with Cursor’s additional training.
Why the Silence?
Cursor did not initially disclose its reliance on Kimi, which raises questions about transparency. Beyond potential embarrassment over not developing a model from scratch, there’s a geopolitical dimension. The AI industry is increasingly framed as a competition between the U.S. and China, and some American companies may be reluctant to publicize their dependence on Chinese technology.
This situation highlights the complex reality of AI development, where even highly-valued startups often rely on pre-existing open-source work, sometimes from competitors in other nations.
Ultimately, this case underscores how much the AI “arms race” depends on collaboration and shared foundations, even as national interests push for self-sufficiency.
