Cursor’s Composer 2 Built on Kimi Model, Company Confirms After Backlash

Cursor’s release of its new Composer 2 coding model is drawing scrutiny after the company acknowledged it was built in part on Moonshot AI’s open-source Kimi model, raising questions about transparency and the evolving structure of AI model development.

The U.S.-based startup, valued at $29.3 billion, introduced Composer 2 as a model delivering “frontier-level coding intelligence.” Shortly after the launch, a user on X claimed the system relied on Kimi, an open-source model developed by Chinese company Moonshot AI. Cursor later confirmed the connection, stating that roughly a quarter of the compute used in the final model came from the Kimi base, with the remainder derived from its own training.

Company executives said the approach complied with licensing terms, and Moonshot’s Kimi account publicly endorsed the integration, describing it as part of an “authorized commercial partnership” with Fireworks AI. Cursor’s vice president of developer education said the model’s performance differs meaningfully from Kimi due to additional training, while co-founder Aman Sanger acknowledged, “It was a miss to not mention the Kimi base in our blog from the start.”

The episode highlights how AI companies are increasingly building on top of open-source foundations rather than training models entirely from scratch. Composer 2 is positioned as a Cursor-specific system, integrated into its coding environment and tuned for agent-based workflows, rather than a broadly distributed standalone model.

At the same time, the situation underscores the importance of attribution as model development becomes more layered. The omission of Kimi from the initial announcement drew attention not only because of the technical dependency, but also because Moonshot is a Chinese firm, adding sensitivity amid ongoing U.S.-China competition in AI.

Cursor has emphasized that its contribution lies in additional training and integration within its platform, pointing to differences in benchmark performance and workflow capabilities. The company’s framing reflects a broader shift in the industry, where competitive positioning is increasingly tied to how models are applied and deployed, not just how they are trained.

The use of an open-source base also reflects the economics of modern AI development. Training large models from scratch requires significant compute and capital, leading some companies to build on existing systems and focus resources on fine-tuning and product integration.

For developers, the practical impact may be limited to how the model performs inside Cursor’s environment. For the industry, however, the incident offers a clearer view into how new AI systems are being assembled—and the trade-offs between speed, cost, and transparency that come with that approach.

This analysis is based on reporting from National Today.

Image courtesy of Cursor.

This article was generated with AI assistance and reviewed for accuracy and quality.