Южная Корея строит собственный ИИ: 519 миллиардов параметров

Some numbers seem meaningless at first glance, but they represent something truly substantial. 519 billion parameters isn't just a technical specification of a language model. For South Korea, this figure represents an attempt to answer the question: can a country have its own, independent artificial intelligence?

SK Telecom has passed the first evaluation phase of a government project to create a sovereign base AI model. In simpler terms, South Korea has launched a national program to develop its own foundational AI, and SK Telecom has become one of the project's key participants.

Что такое суверенный ИИ и его необходимость

What is “Sovereign AI” and Why is it Needed?

Today, most powerful language models are created by a few major players – predominantly American companies. Countries wishing to use AI in government systems, healthcare, education, or critical infrastructure must either rely on foreign developments or build something of their own.

“Of their own” in this context means more than just adapting an existing model to the local language. It refers to a model built from scratch for national needs, trained on data that remains within the country, and capable of operating in sensitive sectors without depending on external suppliers. This is precisely what is called sovereign AI.

South Korea is not the first country to move in this direction, but the project's scale and its government backing make it a notable event in the global AI race.

Особенности модели ИИ и значение 519 миллиардов параметров

What Kind of Model is It, and Why is 519 Billion a Big Deal?

The model developed by the SK Telecom team belongs to the class of so-called omnimodal systems. To put it simply, most familiar AI assistants can work with text, images, or sound – but rarely with all of them at an equally high level. An omnimodal model is designed from the outset to perceive and process different data types as a single whole.

Parameters are, roughly speaking, the “settings” within the model that determine how it understands and generates information. The more parameters, the more nuances and patterns the model can capture. 519 billion is a level comparable to some of the most powerful models in the world. For a national project still in its first phase, this is a significant statement.

Первая фаза проекта: что она означает на практике

The First Phase is Complete – What Does This Mean in Practice?

Completing the first evaluation stage is not the finish line, but rather a confirmation that the team is moving in the right direction and can proceed to the second phase. In government projects of this scale, such “checkpoints” aren't just for show. They signify that independent assessors have reviewed the results and deemed them sufficient to continue.

For SK Telecom, this is an important signal: the technical foundation is in place, the approach is working, and trust has been earned. Now, the team must move forward to the more complex tasks of the second phase, where the model's scale and requirements will likely increase.

Как устроена работа над масштабным ИИ проектом

An Inside Look: How a Project Like This is Managed

Behind the numbers are people. The team working on the model within SK Telecom is called the Omnimodal Foundation Model Team – and judging by how they describe their work, this is not a standard corporate development project. It requires balancing deep research with the strict practical requirements of a government program.

Working on a foundational model of this scale is always a multi-layered task: it's not enough to simply train the model; its quality, security, and cultural and linguistic relevance must also be ensured. For the Korean model, this means a deep understanding of the Korean language, local context, and applications – something not easily replicated by relying on Western counterparts.

Значение корейского ИИ-проекта для других стран

Why This Matters Beyond Korea

At first glance, this might seem like the internal affair of a single country. But in a broader context, these developments reflect a clear trend: nations are increasingly investing in creating their own AI systems, unwilling to be completely dependent on a handful of tech giants.

This is changing the industry landscape. While the large language model race previously involved mainly private companies from the US and China, it is now being joined by national programs from various countries – each with its own priorities, languages, and data requirements.

The South Korean project is also interesting because it brings together a major telecom operator and national objectives. SK Telecom is not a startup or a research lab; it's a company with real infrastructure, experience handling large volumes of data, and access to a broad audience. If the model succeeds, it has the potential to become not just an academic achievement but a functional tool in real-world services.

Открытые вопросы по проекту суверенного ИИ

Open Questions

It's still too early to say how the entire project will end. The first phase is complete, but how many phases there are in total, how much more difficult the next stages will be, and when the model might move beyond the research phase – all of this remains to be seen.

It is also unknown whether the model will be publicly available or will remain a tool for government and corporate use. This is a significant question: sovereign AI can be sovereign in different ways – for the entire country or only for its institutions.

But one thing is already clear: 519 billion parameters is more than just a number in a press release. It is a testament to the fact that South Korea views its own AI as a strategic necessity, not an option.