Does a Chinese - skinned large - language model prop up a $50 billion valuation? Let's dig into the "skinning" suspicions of Cursor.

1. This weekend, there was the most eye - catching AI news that shook both sides of the Pacific Ocean, and even Elon Musk paid attention to it.
   Yesterday, the AI programming tool Cursor launched its "own" model, Composer 2.

   

   The above is a screenshot from the official website. It still says "in - house model" when you click in now.
   Since the release of Composer 1 in October 2024, the outside world has always suspected that it is a repackaged Chinese model because its behavior is very similar, but it has been difficult to find evidence.
   Now that Composer 2 has arrived, many people have started to study what model is behind it. Is it really Cursor's own?
   In order to prevent cracking, Cursor has made many restrictions, but there is a slip in a hundred precautions. A foreign X user @fynnso discovered that there is a place that was prohibited in the previous version but can be executed in this version.
   First, set up a server of your own to act as the calling interface for the AI model. It doesn't matter if there is no model, as long as it can receive client requests.
   Then, in the local Cursor, set the model in use to Composer 2, and the model URL is the server you just set up. In this way, Cursor will send requests to your server, so that you can see what model it is requesting.
   The truth was revealed. The model ID it requested turned out to be kimi - k2p5 - rl - 0317 - s515 - fast (see the picture below).

   

2. The foreign X user posted the above screenshot on the Internet. This caused an uproar. It was obvious to anyone that this was iron - clad evidence that Composer 2 is actually a repackaged Kimi K2.5.

   

   可笑的是，事情一爆发，Cursor 第一时间就把漏洞堵上，现在已经没法复现这个请求（下图）。

   

   But it was too late. It spread all over the Internet, and even Elon Musk tweeted: "It's Kimi K2.5".

   

   Now it has become an open secret and can no longer be covered up.

3. Everyone's attention soon shifted to whether Cursor was infringing on copyright. Although Kimi K2.5 is an open - source model, it uses a modified MIT license (see the picture below).

   

   The license says: You can use this model freely. The only condition is that if your commercial product has a monthly active user base of more than 100 million or a monthly revenue of more than 20 million US dollars, you must disclose in a prominent position in the user interface that you are using Kimi K2.5.
   Cursor's latest disclosed annualized revenue is 2 billion US dollars, which is equivalent to a monthly revenue of 167 million US dollars, obviously meeting the above conditions. However, it hid the fact that it used K2.5.
   Just when everyone thought Cursor was infringing on copyright, one of their leaders finally couldn't sit still and came out to speak.

   

   He admitted that they did use Kimi K2.5, but there was no infringement. Their license came from their partner Fireworks AI.
   Later, the official Kimi also tweeted.

   

   The official Kimi confirmed that Cursor obtained authorization from Fireworks AI. The latter is a Chinese - American AI company in Silicon Valley engaged in the fine - tuning and reinforcement learning of AI models. It obtained authorization from Kimi to retrain the model and then sub - licensed it to Cursor.

4. The matter is basically clear here. Cursor did not violate Kimi's licensing terms, so there is no infringement.
   Since that's the case, why did it desperately cover up this fact? Is it so difficult to admit it openly and provide a modified version of the Kimi K2.5 model?
   I guess the reason is related to Cursor's continuously inflated valuation.
   Bloomberg reported this month that Cursor is conducting the next round of financing with a valuation of $50 billion.

   

   Do you know what its previous valuation was?
   In October 2023, when Cursor was founded, its valuation was $50 million; in the Series A financing in August 2024, the valuation rose to $400 million; in the Series B financing in December, the valuation quickly rose to $2.6 billion; in the latest round of financing in November 2025, the valuation had reached $29.3 billion.
   You can see that the valuation will

Double. Such a rocket - like growth rate requires performance support. However, it is just a modified version of VS Code and uses all open - source technologies.

In order to support the increasingly high valuation, it has the motivation to package itself from an AI tool into a large - model company with model R & D capabilities.

I think this is the main reason why it is reluctant to disclose the use of Kimi K2.5.

5.
Throughout the whole incident, Cursor is undoubtedly the loser, while Kimi is the winner this time, getting a large wave of high - value exposure for free.
When Cursor released Composer 2, it disclosed performance and cost comparisons.
The performance of Composer 2 is lower than that of GPT - 5.4 but higher than that of Opus 4.6.

However, its generation speed is faster than both GPT - 5.4 and Opus 4.6, and the cost is also the lowest.

Since Composer 2 is just a fine - tuned Kimi K2.5, then directly using Kimi can also achieve the same effect.

6.
Previously, some people abroad always accused Chinese companies of stealing foreign technologies. However, this incident has proved that Chinese companies also have technology exports. Those foreign star companies are secretly using Chinese technologies behind the scenes.
Recalling that last week, Yang Zhilin, the founder of Kimi, received an invitation from Jensen Huang to give a speech at the Nvidia GTC conference here, and he was the only representative of a Chinese large - model company.

He presented on stage the paper Attention Residuals just published by the Kimi team.

This new technology is said to be able to significantly improve the inference ability of large models.
My thought is that everyone should have confidence in domestic large models and can use them with confidence in daily work. The gap between domestic large models and foreign flagship models is constantly narrowing, and they are also affordable.

According to Yang Zhilin here, the next - to - be - released K3 model will have a huge performance improvement. Even if it is not 10 times stronger, it is much stronger than K2.5. We can look forward to it.

(End)

    <h3>Document Information</h3>

• Copyright notice: Free reprinting - non - commercial - non - derivative - keep attribution (Creative Commons 3.0 license here)
• Publication date: March 21, 2026

This article is auto-translated by AI.