Cursor, the AI coding startup valued at $29.3 billion, quietly built its new Composer 2 model on top of Kimi 2.5, an open source model from Chinese company Moonshot AI. The company didn’t mention this until an X user called them out, TechCrunch AI reports.
Here’s what happened: Cursor launched Composer 2 this week, marketing it as offering “frontier-level coding intelligence.” Sounds impressive. But an X user named Fynn quickly spotted code identifying Kimi as the underlying model. “At least rename the model ID,” Fynn scoffed. Not exactly a sophisticated cover-up.
The Admission
Cursor’s VP of developer education Lee Robinson confirmed the finding: “Yep, Composer 2 started from an open-source base!” He added that only about 25% of the compute spent on the final model came from Kimi’s base, with the rest coming from Cursor’s own training. Robinson argued this makes Composer 2’s benchmark performance “very different” from Kimi’s original results.
Moonshot AI, backed by Alibaba and HongShan (formerly Sequoia China), didn’t seem bothered. The Kimi account on X congratulated Cursor, confirming the use was part of “an authorized commercial partnership” through Fireworks AI. “We are proud to see Kimi-k2.5 provide the foundation,” the account posted.
Why the Secrecy Matters
Cursor co-founder Aman Sanger admitted the quiet part out loud: “It was a miss to not mention the Kimi base in our blog from the start. We’ll fix that for the next model.”
But the real question is why they left it out in the first place. A few things stand out:
- Optics problem. Cursor raised $2.3 billion last fall and reportedly exceeds $2 billion in annualized revenue. Investors and users might expect a company of that size to build its own models from scratch.
- The US-China angle. Building on a Chinese model feels politically sensitive when the AI competition between the two countries keeps getting framed as an existential battle. Remember the panic after DeepSeek dropped its competitive model early last year?
- Transparency gap. Using open source models is perfectly fine. Not disclosing it when you’re charging premium prices for “frontier-level” capabilities is a different story.
What This Signals
This incident highlights something important about the current AI landscape: the line between “building a model” and “fine-tuning someone else’s model” is getting blurry. And companies aren’t always eager to clarify which side they’re on.
It also shows that open source Chinese AI models are becoming foundational infrastructure, even for well-funded American startups. That’s a significant shift from the narrative that US companies need to “win” the AI race by building everything domestically.
For Cursor users, the practical impact is probably minimal. If Composer 2 performs well, the base model’s origin matters less than the results. But for the industry, the transparency question is worth watching. When companies charge enterprise prices, customers deserve to know what’s under the hood.
More details are available in the original report from TechCrunch AI.