Cursor's Composer 2 coding model built on Moonshot AI's Kimi, not developed from scratch
Cursor acknowledged that its newly launched Composer 2 coding model was built on Moonshot AI's open-source Kimi 2.5 model, not developed independently. The admission came after user scrutiny revealed code references to Kimi. Cursor claims only 25% of compute came from the base model, with 75% from its own training and reinforcement learning.
Cursor's newly launched Composer 2 coding model was built on top of Moonshot AI's open-source Kimi 2.5, the company confirmed after user discovery and public scrutiny.
Cursor promoted Composer 2 as offering "frontier-level coding intelligence" in its launch announcement but did not disclose the Kimi base. A user posting under the name Fynn identified code references to Kimi within Composer 2, prompting questions about the model's actual provenance.
The Admission
Cursor's Vice President of Developer Education Lee Robinson subsequently acknowledged on X: "Yep, Composer 2 started from an open-source base!" He characterized Cursor's contribution as substantial, claiming "only ~1/4 of the compute spent on the final model came from the base, the rest is from our training." Robinson argued this additional training produces "very different" benchmark performance compared to Kimi 2.5.
Cursor co-founder Aman Sanger later stated: "It was a miss to not mention the Kimi base in our blog from the start. We'll fix that for the next model."
Licensing and Partnership
Both Cursor and Moonshot AI confirmed the arrangement was legitimate. Moonshot's official X account posted congratulations to Cursor, stating the company used Kimi "as part of an authorized commercial partnership" with Fireworks AI. The account added: "We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support."
Cursor's licensing terms with Kimi comply with the model's open-source agreement, according to both parties.
Context
Cursor is a well-funded U.S. startup that raised a $2.3 billion Series C round last fall at a $29.3 billion valuation and is reportedly generating over $2 billion in annualized revenue. The company positions itself as a leading AI coding assistant competing with GitHub Copilot and other developer tools.
Moonshot AI is a Chinese AI company backed by Alibaba and HongShan (formerly Sequoia China). Kimi is its consumer-facing language model with open-source variants available.
Why the Initial Omission?
The lack of upfront disclosure raises questions about optics and geopolitical sensitivity. Building on a Chinese model's foundation carries potential reputational risks in the current U.S.-China AI competition narrative, particularly following competitive Chinese models like DeepSeek gaining attention in 2024. Sanger's acknowledgment of the "miss" suggests the omission was deliberate rather than accidental.
The incident highlights tensions in the open-source AI ecosystem: companies can legally build on open models but face scrutiny over transparency and attribution practices. Cursor's approach—substantial additional training (75% of compute) but initial non-disclosure—sits in an ethically gray zone that the company now plans to address in future releases.
What This Means
This case demonstrates that even well-funded AI startups sometimes choose expediency over attribution in model development. It also shows open-source AI foundations are actively fueling commercial products, with legal partnerships enabling rapid deployment. The geopolitical dimension suggests U.S. AI companies may face reputational pressure when building on non-Western model bases, even when technically and legally sound. Cursor's commitment to future transparency suggests the industry may be moving toward clearer base-model disclosure standards.
Related Articles
Amazon, Cursor, Microsoft, OpenAI, and Vercel Launch Agent Plugins, a Shared Packaging Standard for AI Agent Extensions
Amazon, Cursor, Microsoft, OpenAI, and Vercel have released Agent Plugins, an open standard defining a single package format for AI agent extensions. Version 1.0.0 covers Agent Skills and MCP servers, but leaves marketplaces, permissions, and runtime out of scope.
NVIDIA Releases Magpie TTS Multilingual Update: 364M-Parameter Open-Weights Model Now Supports 12 Languages, Sub-50ms La
NVIDIA's Magpie TTS Multilingual, a 364M-parameter open-weights text-to-speech model, now supports 12 languages after adding Modern Standard Arabic, Korean, and Brazilian Portuguese. The model achieves 32ms time-to-first-audio on B200 GPUs and improves speech quality across French, Spanish, and German.
Apple Confirms Siri's New AI Features in China Run on Alibaba's Qwen Models, Not Gemini
Apple has confirmed that its new Siri AI features in China will be powered by Alibaba's Qwen models rather than Google Gemini, which handles Siri AI elsewhere. A published guide indicates users will need separate Qwen accounts, suggesting the privacy protections built into Apple's Private Cloud Compute won't apply.
GitHub Models Retired, Cutting Off Free LLM API Access for GitHub Actions
GitHub Models, the unified LLM API and playground that let GitHub Actions workflows make free AI calls using existing GitHub credentials, has been fully retired. GitHub has not disclosed a reason, though rising inference costs from coding-agent usage are the likely driver.
Comments
Loading...