Cursor AI Model: Chinese Roots & Low Cost Coding Revealed

Cursor, a well-funded U.S. Startup valued at $29.3 billion as of November, has confirmed its new coding model, Composer 2, is built upon the foundation of Kimi K2.5, an open-source model developed by Chinese AI company Moonshot AI. The acknowledgement, delivered over the weekend via X (formerly Twitter), came after an X user pointed out code similarities suggesting the connection, a detail Cursor initially omitted from its launch announcement. The disclosure highlights the complex supply chains underpinning even cutting-edge American technology and raises questions about transparency in the rapidly evolving AI landscape.

The revelation surfaced when an X user named Fynn posted on Friday that Composer 2 appeared to be “just Kimi 2.5” with additional reinforcement learning applied. Fynn supported the claim with code snippets that seemed to identify Kimi as the underlying system. Cursor’s vice president of developer education, Lee Robinson, responded by acknowledging the Kimi K2.5 base, stating that roughly a quarter of the computational work for the final model came from the original, with the remaining three-quarters consisting of Cursor’s own training and fine-tuning.

A Cost Advantage in a Competitive Market

Cursor’s Composer 2 is priced at $0.50 per million input tokens and $1.50 per million output tokens, significantly undercutting competitors like Anthropic’s Claude Opus 4.6 ($5.00/$25.00) and OpenAI’s GPT-5.4 ($2.50/$15.00), according to Cursor’s own marketing materials. This aggressive pricing strategy is a key component of Cursor’s value proposition, positioning it as a more affordable option for developers. The company frames this as “a new, optimal combination of intelligence and cost.” The lower price point is particularly notable given the high costs associated with training and running large language models.

The decision to build on an existing model, rather than developing a proprietary one from scratch, is a strategic one. Both Anthropic and OpenAI invest billions in building their own base models, a level of investment Cursor currently cannot match. As Cursor co-founder Aman Sanger admitted on X, “It was a miss to not mention the Kimi base in our blog from the start. We’ll fix that for the next model.” This suggests a recognition that full transparency is crucial, even if it means acknowledging a reliance on external resources.

The Kimi Connection and Commercial Partnership

Moonshot AI, the developer of Kimi K2.5, is a Chinese startup backed by Alibaba and HongShan (formerly Sequoia China). The company’s Kimi model has gained recognition for its performance, and Cursor’s decision to leverage it reflects a pragmatic approach to accessing advanced AI capabilities. Moonshot AI itself acknowledged the partnership on X, stating that Cursor is using Kimi K2.5 under an “authorized commercial partnership” with Fireworks AI, an inference provider. This partnership appears to address any potential licensing concerns. TechCrunch reports that Cursor maintains the use of Kimi is consistent with the terms of its license.

The choice to utilize a Chinese-developed model is noteworthy, particularly given the current geopolitical climate and increasing scrutiny of technology supply chains. Building on top of a Chinese model, as Business Insider points out, “feels particularly fraught right now.” However, Cursor’s emphasis on its own substantial investment in training and fine-tuning – representing roughly 75% of the total compute – suggests an attempt to differentiate its offering and mitigate potential concerns about reliance on a foreign-developed base model.

Performance and User Reaction

Despite the initial lack of disclosure, the performance of Composer 2 has garnered positive attention. Cursor claims the model scores 61.3 on its internal CursorBench, a significant improvement over its predecessor, Composer 1.5 (44.2), and competitive with Claude Opus 4.6 (58.2) and GPT-5.4 (63.9). Some users on X have even praised Kimi’s performance after learning of its role in Composer 2’s development. One user commented that seeing Kimi 2.5 outperform Claude Opus 4.6 on coding benchmarks was “wild.”

However, not all reactions have been positive. Some critics question Cursor’s initial decision to remain silent about the Kimi base model, suggesting a lack of transparency. One X user, going by the name aira, characterized Cursor as becoming “a model routing layer, not an IDE,” implying the company is simply selecting the cheapest available model rather than innovating independently.

Implications for the AI Coding Landscape

Cursor’s approach highlights a growing trend in the AI industry: leveraging existing open-source models as a foundation for building specialized applications. This strategy allows companies to reduce development costs and accelerate time to market. However, it also raises questions about intellectual property, transparency, and the potential for supply chain vulnerabilities. The incident underscores the importance of clear communication and responsible sourcing in the AI ecosystem. The Decoder notes that Cursor’s silence stemmed from a fear of conceding it couldn’t build a frontier model on its own.

What’s Next for Cursor and Composer 2?

Cursor has stated its intention to conduct “full pretraining” in the future, suggesting a long-term goal of developing its own proprietary base model. In the meantime, the company will continue to refine and improve Composer 2, leveraging user feedback and ongoing training. The company’s success will likely depend on its ability to maintain a competitive edge in pricing and performance while addressing concerns about transparency and supply chain security. The next model release will reportedly include a disclosure of the base model used, addressing the criticism surrounding the initial launch of Composer 2.