Cursor recently revealed that its new coding AI model, Composer 2, is based on the Chinese open-source model Kimi K2.5 developed by the startup Moonshot AI. Initially, the company did not disclose this origin, but the connection came to light after a user identified code pointing to Kimi’s involvement. Composer 2 leverages Kimi as a foundational base and adds substantial reinforcement learning training on top.
Cursor’s cofounder Aman Sanger admitted on social media that omitting mention of Kimi K2.5 in the initial announcement was an oversight. The model was chosen because it scored highly in perplexity-based evaluations among many base models Cursor tested. The company emphasized that although Kimi serves as the starting point, the majority of compute spent on Composer 2’s final form came from Cursor’s own training processes.
This collaboration is authorized through a commercial partnership, with Moonshot AI actively supporting Cursor’s continued pretraining efforts. That partnership reflects a growing trend where companies blend open-source foundations with proprietary enhancements to accelerate AI development.
Composer 2 coding AI pricing and competitive advantages
Cursor touts Composer 2 as a high-level coding AI priced significantly below major rivals. At $0.50 per million input tokens and $2.50 per million output tokens, it undercuts competitors like Anthropic’s Claude Opus 4.6 and Claude Sonnet 4.6, which cost multiple times more. This aggressive pricing could disrupt the AI coding services segment by making advanced capabilities accessible at a fraction of the cost.
- Composer 2: $0.50 per million input tokens, $2.50 per million output tokens
- Claude Opus 4.6: $5 per million input tokens, $25 per million output tokens
- Claude Sonnet 4.6: $3 per million input tokens, $15 per million output tokens
The revelation that a Chinese-developed model underpins Composer 2 has sparked mixed reactions online. Some applaud the performance gains and see it as a sign that reinforcement learning methods from Chinese AI are advancing strongly. Others criticized Cursor for the delayed transparency and suggested the firm functions increasingly as a model integrator, selecting the most cost-efficient base model and layering user-facing features rather than creating models entirely in-house.
As the AI coding tool sector grows more competitive, Cursor’s strategy of combining open-source innovation with its own training and providing access at a sharp discount may prove influential. It also highlights the complex supply chains behind modern AI models, where multiple global players contribute to end products.