# This is Kimi K3, the Chinese AI you can install on your computer

> This is how Kimi K3 works, the Chinese AI you can download onto your computer and which challenges the dominance of OpenAI and Google at the local level
> By Javier Callejo · AI Agents · 2026-08-02

Kimi is the artificial intelligence (AI) model from the Beijing-based company Moonshot, known for being one of the country's six AI tigers. Known as Kimi 3K (official designation), Kimi 3V (a designation following software industry tradition) or simply Kimi, it has caused a particular stir in an industry, AI, that never stops churning and where "revolutions" follow one another at speed. In this field, the experiments of what was already considered a revolutionary change have barely landed - as has happened with Claude Mythos 5, for example - and we already have something presented as having the characteristics of inaugurating a new era. But here eras seem to shrink to a couple of weeks. The next revolution will quickly follow.

## The Chinese AI Kimi K3

There is no doubt that Kimi represents a significant change in the AI industry. It is not just another chatbot. Note that we are already at a stage where chatbots seem like child's play to us. Well, Kimi is considerably more than that.

Moonshot has launched [Kimi K3](https://www.kimi.com/es-419) as an *open-weight*model. That is, as a model that developers can download and implement, subject to its license. This sets it apart from fully closed or proprietary models, such as ChatGPT or Claude. It is a model whose numerical parameters - the final result of its training that defines how it works - are publicly released so that anyone can download, modify and run them. The key points are access to the parameters, local execution and customization.

In this way, the exact fixed weights and values that the model learned and stored during its training phase are shared. The fixed values are exact numbers that are recorded when the AI finishes training and no longer change on their own. Thus, they define the "memory" and the logic of the system, indicating the importance given to each piece of data received. As a result of machine learning, they represent everything the model managed to absorb through testing and error correction on its original servers.

## *Open-weight* is a relative form of openness

Because these numerical parameters are public, anyone can copy those files. They can be downloaded from the system so that they run on users' own computers or servers, without the need to use a given company's cloud. It also makes it possible for other creators or developers to modify or adapt those numbers to improve the system or teach it new tasks.

Put that way, it sounds like an echo of *open-source* or open code. But no. At least, *open-weight* is not full *open-source* . It means you have the file of the AI's "brain"; but not the exact data used to train it, nor, above all, the original code of that training. It does not include the training code: the scripts or frameworks used to create and manage the training dataset. *Open-weight* is a relative form of openness. You can go in, but it is a space with closed-off areas.

In any case, such relative openness makes practical sense if the model's capabilities are at least similar to those of the most effective AI models for users. In this respect, Kimi appears to be up to the task in areas such as coding, reasoning, long-context comprehension and agent workflows, although the most powerful proprietary or closed models still lead on some benchmarks.

## Its Western competitors

Perhaps one of Kimi's main advantages is that it allows work, within a single conversation, with large collections of documents, extensive codebases or entire books. In addition, according to statements from Moonshot itself, architectural improvements have been introduced in order to achieve more solid performance. Advantages or added values that, as with cars, use over a period of time will put to the test.

It will be users who say whether Kimi is the best AI model or whether there are better ones. That said, it is a source of competitive pressure for the main producers of [AI models](https://observatorioblockchain.com/ia/en-diez-anos-los-ordenadores-con-ia-costaran-como-un-smartphone/): OpenAI, Anthropic, Google or Meta. For now, it has become a particular draw for developers. And Kimi may not be important in itself, for the practical advantages and functions it puts in the hands of the end user, but for what it represents.

Being a relatively open model (*open-weight*), anyone with enough computing resources can run and modify it. This makes it possible to speed up innovation processes even further. Doing so, moreover, at a considerably lower cost than proprietary models, which require enormous investment. Therefore, Kimi K3 not only shows that Chinese AI companies are more advanced than was believed, but its own *open-weight*system, which makes the model a vessel for continuous modification, becomes a springboard for reducing the technological advantage that US AI companies hold.

## AI never stops being revolutionized

The potential threat that the Kimi K3 model may pose has put its Western competitors on alert. The first reactions have come in the rhetoric of cybersecurity, pointing out that Kimi K3 is less powerful than their models at responding to cyberattacks. In other words, they have started by labeling it vulnerable. At least, more vulnerable to cyber risk. On the other hand, the fact that it is a Chinese company fuels the political debate.

Kimi K3 once again makes clear that what is revolutionary about AI is that it never stops being revolutionized. And that it will not stop giving us revolutions periodically.
