Sber has released Europe’s largest open source foundation models under the MIT licence.
Sber has unveiled Europe’s largest open source AI release, opening model weights and code for a full suite of Russian-trained neural networks under the MIT licence. The move is designed to strengthen national technological sovereignty by enabling corporations, startups, and government-linked entities to deploy and fine-tune AI capabilities securely within private infrastructure while retaining full data control.
“We believe creating world-class artificial intelligence requires two things: massive resources and world class R&D teams. Sber has both. But what matters most is sharing, not locking down technology. Our strategy is to become an open foundation for innovation nationwide. That is why we are releasing model weights. This is a pivotal moment.
Any company in Russia, whether a bank or startup, can install these models within their closed systems, fine-tune them on sensitive internal datasets, and retain complete control over their confidential information.
This approach reflects true technological sovereignty: AI belongs to the entire nation, driving business transformations and economic growth. I would also like to note that Ultra will be soon available for corporate clients, with optimized cost of ownership for internal corporate deployments.”— Andrey Belevtsev, Senior Vice President, Head of Technology & AI, Sberbank.
The flagship additions include GigaChat Ultra Preview and GigaChat Lightning, two Mixture-of-Experts language models excelling in Russian-language performance. Ultra Preview already surpasses DeepSeek V3.1 on benchmark quality metrics, while Lightning enables fast, local execution on laptops.
Sber has also open sourced GigaAM-v3 speech recognition models, outperforming OpenAI Whisper for Russian speech, alongside Kandinsky 5.0 for high-quality image and video generation with native Cyrillic capabilities. The powerful K-VAE 1.0 visual encoders further enhance generative performance and scalability.
