Mishustin cites Alisa traffic lead over foreign apps and existing safety rules ahead of industrial rollout
Russian foundational models trained on local-language data are already in wide use at home, Prime Minister Mikhail Mishustin said in remarks the government posted Sept. 25.
The models work, he said, because they pick up what users actually mean. Local context. Linguistic quirks. Cultural habits. The country’s own texture.
“Russia possesses such models, and they are of high quality,” Mishustin said. “They are widely used by our citizens because they effectively “understand” users – the local context, linguistic nuances, cultural features, and the broader characteristics of the country and region. This is largely due to their training on Russian-language datasets, which incorporate these specific elements.”
One service sits at the center of the claim. The voice assistant Alisa and the generative model behind it.
“One example is the voice assistant Alisa and its generative model. In terms of traffic, this service has managed to outperform all foreign AI applications available in Russia.”
That comparison is the number officials are putting forward. Traffic, not benchmark scores. Alisa ahead of the foreign apps still reachable inside the country.
Its developers, Mishustin said, built products people actually use. Popular. Convenient. And, in his telling, usable beyond Russia’s borders.
“Its developers have created products that are both popular and convenient for users, with potential applicability across the Eurasian Economic Union.”
Partners in the union, he added, can take an open model or a version tuned to national conditions, with Russian engineering support attached.
“As our valued partners, you have the opportunity not only to download an open model, but also to obtain a solution tailored to national specifics, with Russian engineering support. This approach enables further refinement and improvement of these products.”
Neural networks, he said, have already moved from novelty to daily tool. Work and home. The point he kept returning to was access.
“Today, neural networks are becoming true assistants in both professional settings and everyday life, and it is important that they remain accessible to a broad range of users.”
The same remarks tied the models to a larger industrial push. AI is already changing civilian unmanned systems and transport, Mishustin said. The legal piece, he added, came first.
“In Russia, under the President’s instruction, the necessary regulatory framework has been established to support the development of this field, ensuring the safety of emerging technologies before their transition to commercial and industrial use.”
That sequence is the part executives and developers have to weigh. Safety rules on the books before the handoff to commercial and factory-scale use. Not after.
The speech sat inside a wider pitch on digital industry and Eurasian cooperation. IT’s share of Russian GDP, Mishustin said, doubled over six years to 2.7 percent. Software and related services topped 5 trillion rubles and grew more than fourfold. Sales of Russian software and development services hit 2.5 trillion rubles last year, up 20 percent. The industrial software market reached about 50 billion rubles.
Those figures sit in the same transcript as the Alisa claim. They are the backdrop officials are using: domestic software replacing departed foreign tools, then generative models trained on Russian data taking traffic.
For teams building or buying models aimed at Russian-speaking users, the official line is now public. Local training data is presented as the reason the systems “understand” users. Alisa is the cited traffic winner against foreign apps still available in the market. A regulatory frame for safety is described as already in place, and it is described as preceding commercial and industrial deployment.
The government posting does not give Alisa’s exact traffic figures, model sizes, or training-token counts. It does not name the developers beyond the product. It does state that Russia has foundational models, that they are in wide domestic use, that they were trained on Russian-language datasets, and that Alisa’s traffic has beaten the foreign alternatives on offer inside the country.