This week, the U.S. AI coding startup Cursor, known for its high valuation, officially launched its new model, Composer2, which the company claims possesses "cutting-edge coding intelligence." However, the model quickly sparked a public controversy. On the social platform X, user Fynn pointed out through code evidence that the underlying model of Composer2 appears to be the open-source model Kimi k2.5 recently released by Moonshot AI (Dark Side of the Moon), and questioned whether Cursor directly added reinforcement learning features without renaming the model ID.

Code Internet Computer

Responding to the allegations, Lee Robinson, the vice president of education at Cursor, quickly addressed the issue, acknowledging that Composer2 was indeed developed based on an open-source project. However, he emphasized that only about a quarter of the final model's computation came from the open-source base, with the rest derived from Cursor's own training, which led to significant differences in performance between Composer2 and the original Kimi model. Subsequently, Moonshot AI confirmed this commercial licensing collaboration and stated that their Kimi-k2.5 model was effectively integrated through Cursor's continuous pre-training and high-computing power reinforcement learning, complying with the terms of the open model ecosystem.

Although Cursor co-founder Aman Sanger eventually admitted that failing to mention the Kimi base in the initial announcement was a "oversight," the incident once again triggered discussions in Silicon Valley about "technical purity" and global supply chain collaboration. In the context of escalating global AI arms race, top U.S. startups choosing Chinese high-performance open-source bases for secondary development reflects the technical logic of cross-border flow within the open-source ecosystem, as well as the delicate balance between brand storytelling and supply chain transparency for tech companies.