So Cursor got caught — and it's actually a more interesting story than it first looks
Okay so this blew up yesterday. Cursor — the AI coding tool that Indian developers have been going crazy over — launched a new model called Composer 2 this week. They called it "frontier-level coding intelligence." Big claim, right? Except they conveniently forgot to mention that it's built on top of Kimi 2.5 — an open-source model from Moonshot AI, a Chinese company backed by Alibaba.
Someone on X named Fynn spotted it in the code, called it out publicly, and Cursor had to come clean. Let me walk you through what actually happened — because there are multiple interesting layers here.
Who is Cursor and why do developers care?
If you're an Indian developer who hasn't tried Cursor yet — it's an AI-powered code editor that's been genuinely eating into VS Code's dominance. It uses AI to help you write, debug, and refactor code in real time. It's become incredibly popular with Indian developers in Bangalore, Hyderabad, and Chennai who are building startups or working at tech companies.
Cursor is not a small startup. They raised a $2.3 billion funding round last fall at a $29.3 billion valuation. They're reportedly exceeding $2 billion in annualized revenue. This is a serious, well-funded company — which makes the Kimi situation all the more surprising.
What is Kimi 2.5 and who made it?
Kimi 2.5 is an open-source AI model released by Moonshot AI — a Chinese AI company backed by Alibaba and HongShan (formerly Sequoia China). "Open source" here means the model weights are publicly available — anyone can download and build on top of them, as long as they follow the license terms.
Moonshot AI is part of the wave of Chinese AI labs that have been releasing genuinely capable open-source models. DeepSeek got all the attention earlier this year, but Kimi is in the same tier — strong coding performance, open weights, free to use and fine-tune.
What exactly happened
Cursor launched Composer 2, promoted it heavily, and said nothing about Kimi or Moonshot AI in the announcement. Then an X user named Fynn dug into the model's code and found identifiers pointing to Kimi as the underlying model. The X post was blunt: "Composer 2 is just Kimi 2.5 with some additional reinforcement learning. At least rename the model ID."
Ouch. That last line is the kind of thing that makes PR teams sweat.
Cursor's VP of Developer Education Lee Robinson quickly responded and acknowledged it: "Yep, Composer 2 started from an open-source base!" But he pushed back on the framing — only about one quarter of the compute used to build the final model came from the Kimi base. The remaining three quarters was Cursor's own training. He said Composer 2's benchmark performance is "very different" from Kimi 2.5 as a result.
Moonshot AI's own X account then chimed in and congratulated Cursor, saying Cursor used Kimi "as part of an authorized commercial partnership" with Fireworks AI. They called it "the open model ecosystem we love to support."
So did Cursor do anything wrong?
Technically, probably not. Open-source models exist specifically so companies can build on them. That's the whole point. Cursor followed the license terms, had a commercial partnership arrangement, and added substantial training on top. The Kimi team themselves are happy about it.
The real issue is transparency — or the lack of it. Cursor launched Composer 2 without disclosing that it was built on a Chinese open-source model. In the current geopolitical climate, where US-China AI tensions are real and companies are being scrutinized for their AI supply chains, that's a sensitive omission.
As TechCrunch put it: building on top of a Chinese model "might feel particularly fraught right now, with the so-called AI 'arms race' often framed" in national security terms. It's not that Cursor broke rules — it's that they chose not to disclose something that, once revealed, looked like they were hiding it.
What this means for Indian developers using Cursor
Functionally, nothing changes. If you're using Cursor in Bangalore or Chennai to build your next app, Composer 2 still does what it does. The benchmark performance difference between Kimi 2.5 base and Cursor's trained version is real — they're not the same model.
But this story is a useful reminder that "proprietary" AI tools are often built on open-source foundations. When Cursor, or any other AI coding tool, launches a "new" model — it doesn't always mean they trained it from scratch. Many of the AI tools Indian developers pay monthly subscriptions for are fine-tuned versions of base models from Google, Meta, Mistral, or — now we know — Moonshot AI.
Cursor's pricing: the free tier gets limited usage, the Pro plan is $20/month (roughly ₹1,670/month at current exchange), and Business is $40/month (₹3,340/month). For Indian developers, that's a meaningful spend, and knowing what you're actually paying for is fair.
The bigger picture — Chinese AI models are everywhere
This story accidentally highlights something important: Chinese AI models are deeply embedded in the global AI toolchain. DeepSeek, Kimi, Qwen from Alibaba — these are legitimate, capable models that Western companies are building on. The US government is concerned about this, but practically speaking, if you're banning open-source Chinese AI models, you're also restricting what American companies can build.
For Indian developers: this is actually good news. More open-source models from more places means more competition, better models, and lower costs for anyone building AI products. India's developer community benefits when the AI ecosystem isn't a US-only oligopoly.
My honest take
The not-disclosing-Kimi part was a mistake. Cursor knew they were building on a Chinese model in a politically sensitive environment — the right move was to mention it upfront. "We built Composer 2 starting from Kimi 2.5 as a base, with substantial additional training" is an honest statement that positions it well. Instead, they said nothing, got caught, and now the story is about the cover-up rather than the model itself.
That said: one quarter base model, three quarters original training is actually a meaningful amount of work. Cursor isn't just reselling Kimi — they did real engineering. The benchmarks backing that up matter.
The irony is that if they'd been upfront, this would have been a non-story. "AI company builds on open-source Chinese model" is not shocking in 2026. Getting caught hiding it is.
Comments (0)
Be the first to comment!