In the early hours of March 20, AI programming tool Cursor (parent company Anysphere, latest valuation $29.3 billion) released its in-house model Composer 2, stating in a blog post that the performance improvement came from "continuously pretraining the base model for the first time, combined with reinforcement learning," without mentioning the base model's source.
In less than two hours, developer @fynnso, while debugging a Cursor API request, intercepted the actual model ID of Composer 2: `kimi-k2p5-rl-0317-s515-fast`, literally meaning "Kimi K2.5 + RL." Dark Side of the Moon pre-training lead Du Yulun promptly tweeted, stating that the team tested Composer 2's tokenizer and found it to be "identical to our Kimi tokenizer," almost confirming that "this is the result of further training on our model," and directly questioned Cursor's co-founder Michael Truell: "Why did you not respect our license and did not pay any fees?"
The tweet was later deleted. The controversy quickly escalated on social media, with Elon Musk replying to @fynnso's post, saying, "Yeah, it's Kimi 2.5," further fueling the discussion.
Kimi K2.5 is released under a modified MIT license, explicitly stating that commercial products with over 1 billion monthly active users or monthly revenue exceeding $20 million must prominently display "Kimi K2.5" in the user interface. Given Cursor's valuation and the scale of its paying users, the monthly revenue threshold is almost certainly triggered.
Then the wind shifted. The official account of Dark Side of the Moon @Kimi_Moonshot posted this morning, changing the tone from accusation to congratulations: congrats to the Cursor team on the release of Composer 2, stating "we are proud to see Kimi K2.5 provided the foundation." The statement also clarified that Cursor's access to Kimi K2.5 through Fireworks AI's hosted RL and inference platform is part of an authorized commercial collaboration, with license compliance ensured through Fireworks AI's commercial agreements.
After Kimi's official statement, Cursor co-founder Aman Sanger and VP of Developer Education Lee Robinson followed up. Sanger explained the technical choices: the team conducted confusion matrix evaluations on multiple base models, and Kimi K2.5 "proved to be the strongest," followed by additional pretraining and a 4x scale of high-performance reinforcement learning, deployed through Fireworks AI's inference and RL samplers.
Robinson added that the final model leveraged about 1/4 of its compute from Kimi, with the remaining 3/4 coming from Cursor's own training. Both acknowledged that not mentioning the Kimi base in the blog post at the time of release was a "mistake" and stated that the next model would be promptly disclosed.
This marks Cursor's second instance of utilizing a Chinese open-source model without disclosure. In November 2025, when Composer 1 was released, the community discovered that its tokenizer matched that of DeepSeek, with the model occasionally generating Chinese text during inference, yet Cursor did not provide any clarification at that time.
The discussion sparked by this incident has transcended mere license compliance. Clément Delangue, co-founder and CEO of Hugging Face, commented that this is further validation of Chinese open source, stating, "Chinese open source is now the most significant force shaping the global AI technology stack," emphasizing that the forefront of competition lies not only in who can train from scratch but also in who can adapt, fine-tune, and productize most swiftly.
An intriguing temporal coincidence: On March 15, Bloomberg reported that Moonshot is seeking up to $1 billion in new funding, valuing the company at around $18 billion, more than quadrupling its value from three months prior, with investments from Alibaba and Tencent. Just five days later, the world's highest-valued AI programming tool was found to be based on Kimi K2.5. Anysphere, valued at $29.3 billion, identified Kimi K2.5 as the "strongest base" in evaluations, upon which they have built their most core product, perhaps providing the most direct market endorsement of Moonshot's technological capabilities.
At this juncture, with the fundraising round not yet complete, the Cursor incident effectively served as a global developer-facing showcase of Kimi's capabilities. It raises the question of whether Moonshot's $18 billion valuation still underestimates its potential, warranting a reassessment.
