Cursor Admits New Model Composer 2 Was Developed Based on Domestic Large Model Kimi

After the Compose 2 launch that caused a stir in the industry this week, the top AI coding tool Cursor found itself in a public controversy over "originality." The incident started when a social media user provided code evidence pointing out that this new model, which claims to have "peak-level programming intelligence," is actually based on the open-source model Kimi 2.5 recently released by Moonshot AI.

In response to the accusations, Lee Robinson, the vice president of education at Cursor, quickly replied, admitting that Composer 2 was indeed built on an open-source base. This statement quickly reversed the previous impression of "pure self-research," making the penetration of domestic large models into international top tools a focal point.

Is it "copying" or "deep transformation"?

In response to external doubts about its high valuation, Cursor provided a detailed technical explanation. Robinson pointed out that although Kimi was used as a base, only about one-quarter of the computing power was spent on the base, while the remaining three-quarters were used for Cursor's own pre-training and intensive reinforcement learning (RL).

This means that Composer 2 has significantly outperformed the original Kimi in various benchmark tests. Cursor is trying to prove with this data that although the "foundation" comes from China, the "programming soul" of the upper structure was injected by Cursor.

A compliant business partner, not "learning by stealing"

It is worth noting that this controversy did not escalate into a legal dispute. The official account of Moonshot AI (Kimi) later publicly congratulated Cursor on X and confirmed that both sides had reached a licensing commercial partnership through Fireworks AI. The Kimi team stated that they were very proud to see their model serve as a base, continuously trained by Cursor to create value for global developers.

Since it is a compliant collaboration, why did Cursor not mention Kimi at all in the initial announcement? Cursor co-founder Aman Sanger admitted it was a "oversight" and promised to correct it in the next model.

However, industry analysts generally believe that this reflects the sensitive "arms race" mentality in the current AI field. As a U.S. star startup valued at nearly $3 billion and with annual revenue exceeding $200 million, admitting that its core product relies on a Chinese model obviously carries some psychological pressure in the current political and market environment.