Sber is Russia’s largest bank and one of the country’s leading technology companies, whose ecosystem of AI-powered digital, financial, and retail services reaches tens of millions of users. Anton Frolov heads the Generative AI Development division and oversees the research and development of GigaChat, Russia’s own neural network. In this TNGlobal Q&A, Frolov speaks about why Sber chose to build its own model from scratch, how AI services are used in Russia, and what he sees as the prospects for cooperation with China.
It’s no secret that many Russian companies use Chinese open-source models in their own development. How do you assess the contribution of Chinese AI companies to global technological progress?
Chinese companies have made open source genuinely competitive: the gap between open and proprietary models is narrowing rapidly, and this is no longer a conversation about trade-offs between accessibility and quality. The launch of DeepSeek showed that a powerful model doesn’t require trillion-dollar budgets or access to enormous computing resources. Qwen, DeepSeek, the recently released Kimi, and other open Chinese models have made AI an accessible tool for thousands of teams around the world, including Russian ones.
The more companies that gain access to a quality foundation, the more diverse solutions, specializations, and applications emerge — the market as a whole grows faster than if the technology were confined to a handful of closed labs. China has introduced a different development philosophy into this space: maximum efficiency with limited resources. At Sber, we follow the same principles — but we don’t build on open-source foundations. We train our models from scratch, on our own pre-training. Without control over foundational training, you have no control over the final product.
Building your own model from scratch is expensive. Why not use open source, as many companies do?
Creating AI involves several technological stages. One of them — pre-training — is the most resource-intensive. If you want to build something that is truly your own, control over that stage is critical. It’s roughly analogous to having the source code of a program versus a compiled binary. You can probably, with great effort, modify a pre-compiled program, but it’s very hard and risky. Those who take open-source models and try to adapt them are actually spending enormous effort: when you take someone else’s pre-trained model and try to adapt it to your needs, you break its core foundational knowledge. Not to mention that access to open models can be cut off at any moment, or developers may simply decide not to release a new version.
At the same time, we believe in openness and systematically release models as open source, because we believe Russian AI should become part of the global technology stack. The more teams build on your foundation, the more you cultivate an ecosystem around yourself rather than compete with it. The experience of Chinese developers illustrates this clearly: openness doesn’t kill competitive advantage — it scales influence.
Based on your experience with GigaChat, what is an AI assistant actually for — what is its core value for an ordinary person? Is there anything specific about the Russian audience?
What does a user need? Relief from routine, and an advisor who helps make life decisions — including very difficult ones. Today, people have dozens of apps installed on their phones for different tasks. A universal AI assistant needs to replace all of them. Sber’s mission is to make GigaChat that kind of assistant — one that handles any task a user has, whether short-term (“book a table” or “buy tickets”) or long-term. It needs to be able both to act and to function as an advisor. You simply set a task and the assistant handles it in the background while you go about your life. But that’s just the beginning. The deeper value lies in helping with truly important decisions: health, finances, career.
One of the main barriers today is trust. Research shows that up to 25% of Russia’s working-age population uses AI assistants, but many approach the technology with skepticism. The Edelman Trust Barometer shows that among those who reject AI, only 18% had genuinely negative experiences — meaning the distrust is more a matter of prejudice than conscious rejection. Trust grows through experience: those who have been helped by AI at work report trust levels 26–46 percentage points higher. Our task is to lower the entry threshold so that skepticism gives way to habit.
How will everyday life change over the next five years in terms of how people interact with technology?
In three to five years, a personal AI assistant will become as natural a part of people’s lives as a mobile phone or a bank account. The era of dozens of apps for different tasks is ending — in their place will be a single assistant that knows the user, adapts to their life, and is even capable of changing its own interface. The key will not just be the quality of responses, but autonomy: if a barber calls to cancel an appointment, the assistant will independently rearrange the user’s plans based on the new circumstances. Such an assistant could have an infinite range of capabilities: making a translation, buying tickets, managing a calendar. It is through these capabilities that traditional apps should be replaced.
According to Gartner’s forecast, by 2028 a third of all user interactions will shift from conventional apps to agent-based interfaces. We are already taking steps in this direction: the updated GigaChat AI assistant now features long-term memory — it remembers the user’s interests, profession, and habits between sessions and uses that context in every subsequent conversation. Voice mode has become a full real-time conversation, with automatic internet search, the ability to interrupt the model, and topic changes without losing context. The market will take shape over the next five years, and those who manage to build a truly personal ecosystem around the individual will come out ahead.
What space do you see for Russian-Chinese cooperation in AI?
AI is an industry that by its very nature develops through openness. The industry’s logic is to share architectural solutions, publish models as open source, build on each other’s work, and release optimizations. Progress accelerates precisely when knowledge is not locked inside a single company. Russian-Chinese cooperation should follow this logic. At the R&D level, this could mean joint research programs, knowledge exchange, and co-authored publications — fundamental science works well in an international format, and both sides benefit from open exchange of ideas.