The Kimi 2.5 Controversy: When a $50 Billion Startup Forgot to Credit Its Open‑Source Foundation
A $50B startup's 'breakthrough' model was discovered to be a repackaged, uncredited Chinese open-source AI.
On March 19, 2026, the developer tools startup Cursor announced Composer 2, its latest in-house coding model. The initial reception was overwhelmingly positive, with the company touting a benchmark score of 61.7% on Terminal-Bench 2.0, which notably outperformed Anthropic's flagship Claude Opus 4.6 model (58.0%) at a claimed one-tenth of the cost. The narrative was one of a significant, cost-effective breakthrough in AI-assisted software development.
Within a day, the celebration unraveled. A developer inspecting the API configuration discovered the underlying model identifier: 'kimi-k2p5-rl-0317-s515-fast.' This name directly referenced 'Kimi 2.5,' the open-source model from Chinese AI company Moonshot AI, with suffixes indicating reinforcement learning fine-tuning. The discovery, amplified by a comment from Elon Musk, shifted the conversation from technical achievement to a controversy over transparency and licensing. The core issue became Cursor's failure to credit its foundational model, raising questions about the ethical use of open-source work and highlighting the growing influence of Chinese open-source AI in the global ecosystem.
- Cursor's 'Composer 2' model scored 61.7% on Terminal-Bench 2.0, beating Claude Opus 4.6.
- The model was exposed as a fine-tuned version of Moonshot AI's open-source Kimi 2.5, not an original creation.
- The incident sparked a major debate on transparency, proper licensing, and credit for open-source AI foundations.
Why It Matters
This case sets a critical precedent for ethical AI development, demanding transparency when commercial products are built on open-source foundations.