Cursor Admits Its New Composer 2 Coding Model Was Built on Moonshot AI's Kimi K2.5
The AI coding tool space just got a transparency moment. Cursor, the AI-powered IDE that has quietly become a favorite among professional developers, confirmed that its newly launched Composer 2 model — marketed as delivering "frontier-level coding intelligence" — was built on top of Kimi K2.5, an open-source model from Chinese AI lab Moonshot AI. The admission came after community developers spotted the model ID in API responses before any official disclosure, prompting Cursor VP Lee Robinson to acknowledge that roughly 25% of the model's compute came from the Kimi base, with the rest from Cursor's own proprietary training runs.
The numbers are compelling regardless of the provenance debate. Composer 2 arrives at $0.50/$2.50 per million input/output tokens — an 86% cost reduction versus its predecessor — while benchmarking above Claude Opus 4.6 on Terminal-Bench 2.0, trailing only GPT-5.4. That combination of price and performance is hard to argue with. What makes this story matter beyond the product specs is what it signals about the broader industry: well-capitalized Western AI companies quietly fine-tuning Chinese open-source models under their own branding is becoming a pattern, not an exception. Kimi K2.5's emergence as a serious base model of choice for production products validates Moonshot AI's technical standing in a way that benchmark leaderboards alone cannot.
The incident raises fair questions about transparency in AI product launches. Developers and enterprise buyers increasingly want to understand what's actually powering the tools they depend on — both for security review purposes and to assess supply-chain exposure. As Chinese open-weight releases continue to match or outpace closed frontier models on key benchmarks, expect more product disclosures like this one, voluntary or otherwise.