The government is discussing the possibility of mandating Russian companies and organizations to use domestic artificial intelligence models. According to the draft resolution, if the cost of a sovereign model does not exceed the cost of a national model by more than 1.5 times, preference should be given to it when choosing an AI solution.
The proposal was discussed in late August at a working group on AI regulation. According to "Kommersant", the initiative came from Sber.
It is important to distinguish between two concepts. A sovereign model is considered one that is fully developed by a Russian legal entity and operates exclusively on Russian infrastructure. Currently, only GigaChat from Sber falls under this definition. A national model must also be substantially developed by a Russian company, but its creation may involve the use of individual open-source solutions, such as Alice AI.
The proposed rules may affect more than just businesses. Sovereign or national models, taking into account the established price condition, are proposed for use in the preparation of strategic documents, state programs and national projects, the development of regulatory acts, as well as in the tax, customs, and budget spheres. The list also includes critical information infrastructure, education, healthcare, and public services.
Earlier it was reported that the government is considering the possibility of using domestic AI models also in the field of security, elections, public administration, and education.
The Ministry of Digital Development confirmed that the mandatory use of domestic AI models is indeed being discussed. However, the ministry emphasized that the published list currently represents a set of proposals from working group participants, not a final decision. On a number of initiatives, the positions of the parties still diverge, so it is too early to talk about an approved list of mandatory areas.
Sber also stated that the document contains ideas from various working group participants and is not a proposal from the company itself. The bank believes that the question of where exactly the use of sovereign and national models should be made mandatory requires further elaboration.
Yandex advocated for maintaining competition. The company believes that developers need clear and uniform rules, and the industries themselves should be able to choose technological solutions depending on specific tasks.
MWS AI, part of MTS Web Services, took a similar stance. They consider it premature to limit the list of permissible models in advance. According to the company, such a decision could lead to a situation where for some tasks in critical industries, there simply will not be suitable solutions in terms of quality and cost. At the same time, even in sensitive areas, there are many tasks that are not related to state interests or confidential data.
A separate issue is the price. Danila Egorov, founder and managing partner of the analytical agency AIANA, notes that a sovereign model may turn out to be cheaper than a national one. According to AIANA, 1,000 GigaChat Max tokens cost 0.65 rubles, YandexGPT Pro 5.1 costs 0.8 rubles, while open models cost about 0.2–0.3 rubles.
However, the price of tokens does not show the entire cost of using AI. As Sergey Kobelev, AI advisor at Yakov and Partners, notes, a significant part of the costs comes from integrating the model into workflows and information systems, human verification of its results, and error correction.
Anton Sipachev, Technical Director of the GenAI direction at Just AI, also draws attention to compact models. According to him, solutions such as GigaChat Lightning and YandexGPT Lite can show comparable quality to more powerful models in many applied tasks – for example, when processing requests and classifying documents – with significantly lower operating costs.
At the same time, the expert believes that the main beneficiary of the proposed approach could be the same company, which effectively creates prerequisites for a monopolistic position for it.