On March 19, Cursor officially launched Composer 2, its latest AI coding assistant, with a celebratory blog post. However, just three days later, a developer uncovered a critical detail within the API configuration: the string "kimi-k2p5-rl-0317-s515-fast." This single discovery quickly unraveled a broader narrative concerning transparency, open-source ethics, and the increasingly globalized nature of AI infrastructure.
Investigations revealed that Composer 2 is built upon Moonshot AI's Kimi K2.5, a Chinese open-source Mixture-of-Experts (MoE) model. This finding directly challenged Cursor's prior assertion that "75% of compute was ours," prompting significant discussion within the AI community.
In terms of performance, CursorBench scores positioned Composer 2 at 61.3, though this metric was noted for potentially having a "home-field advantage." More broadly, the Terminal-Bench assessment indicated a gap of merely 3.7 points when compared against Claude, suggesting competitive performance in real-world coding scenarios.
From a cost perspective, Composer 2 presents a compelling advantage. Priced at $0.50 per million input tokens, it is approximately 30 times cheaper than leading models such as Opus 4.6, making advanced AI coding assistance far more accessible.
Interestingly, the adoption patterns among highly productive developers suggest a hybrid approach: many leverage Cursor for approximately 80% of their daily coding tasks, while reserving Claude Code for the remaining 20% of more complex or specialized work.
The Composer 2 controversy underscores the intricate supply chains inherent in AI product development and reignites crucial conversations about vendor transparency and the ethical responsibilities of AI service providers. It also highlights the growing interconnectedness and opportunities within the global AI model ecosystem.