Cursor AI, the popular vibe coding platform, recently hit the headlines for releasing its Composer 2 model to offer what it claimed a superior coding experience than other third-party AI models. Marketed as offering “frontier-level coding intelligence,” a controversy erupted within days of its release, as coding critics discovered that Composer 2 was developed on Moonshot AI’s open-source Kimi K2.5 model as the base. For those who are unaware, Kimi K2.5 model is a Chinese open-source AI model.
As soon as the controversy blew up, Cursor was quick to clarify all concerns, with co-founder Aman Sanger confirming that Composer 2 was based on Kimi K2.5 but trained extensively by Cursor to make it more capable than the stock model. “It was a miss to not mention the Kimi base in our blog from the start. We’ll fix that for the next model,” stated Sanger.
Community discovers Kimi K2.5 base
When Composer 2 was released, the company highlighted impressive results on challenging coding benchmarks, including SWE-bench Multilingual and Terminal-Bench, thus promising a major step forward for multi-file editing, 200,000-token context handling, and complex software engineering tasks.
Within hours, developer Fynn posted findings from API response inspection showing an internal model identifier — “kimi-k2p5-rl-0317-s515-fast” — thus clearly referencing Moonshot AI’s Kimi K2.5, which was originally released in January 2026. Fynn characterised Composer 2 as “essentially Kimi 2.5 plus extra reinforcement learning” and urged Cursor to update the model naming for clarity.
Joining many developers in criticising the discovery was Tesla and SpaceX CEO Elon Musk, who also dropped a comment, stating, “Yeah, it’s Kimi 2.5.”
Cursor issues clarity, co-founder Sanger admits fault
As the incident blew up, Cursor responded promptly. Vice President of Developer Education Lee Robinson acknowledged the open-source foundation, stating, “Yep, Composer 2 started from an open-source base!” He clarified that roughly one-quarter of the total training compute originated from the Kimi weights, while the remaining three-quarters came from Cursor’s own extended training runs. As a result, he said, benchmark scores diverged significantly from those of the unmodified Kimi model.
Co-founder Aman Sanger addressed the communication shortfall, stating, “It was a miss to not mention the Kimi base in our blog from the start. We’ll fix that for the next model.”
Moonshot AI’s official Kimi account welcomed the news on X, confirming an authorised commercial arrangement facilitated through Fireworks AI’s inference platform. “We are proud to see Kimi-k2.5 provide the foundation,” the account posted. “Seeing our model integrated effectively through Cursor’s continued pretraining & high-compute RL training is the open model ecosystem we love to support.”
While some developers praised the collaboration as a healthy example of knowledge sharing across borders, others expressed concern over initial marketing that presented Composer 2 as more independently developed than it was in reality.
Cursor, which recently surpassed $2 billion in annual revenue and secured funding at a $29.3 billion valuation, continues to promote Composer 2 as a transformative tool for professional coding workflows.