MOSCOW, RUSSIA, July 8th, 2026, FinanceWire
Sber today announced the launch of GigaChat 3.5 Ultra, its latest flagship artificial intelligence model designed to deliver stronger performance in coding, mathematics, long-context reasoning, and autonomous AI agent workflows. The new model is now available free of charge through the GigaChat AI assistant for users and as an open-source model for developers building AI-powered applications and services.
The launch represents a significant advancement in Sber's generative AI portfolio, introducing a more efficient architecture that generates long-form text up to four times faster while requiring substantially fewer computing resources than the previous flagship version. Built using Sber's proprietary linear attention architecture, GigaChat 3.5 Ultra is also nearly half the size of its predecessor, enabling broader deployment across enterprise and developer environments.
GigaChat 3.5 Ultra has been designed to support real-world business and technical applications through enhanced code generation, mathematical reasoning, financial calculations, and structured problem solving. According to Sber, the model delivers more accurate, reliable, and clearly organized responses, making it suitable for developers, analysts, engineers, and organizations seeking advanced AI capabilities.
A key feature of the newly launched model is its ability to process and analyze lengthy documents with improved speed and accuracy. Leveraging Sber's proprietary linear attention technology, GigaChat 3.5 Ultra can efficiently work with contracts, technical documentation, reports, and other long-form content by retaining contextual understanding without repeatedly processing the entire document, significantly improving overall performance.
The new model also expands support for autonomous AI agents capable of independently completing multi-step tasks. Users can assign complex workflows that involve gathering information, generating and executing code, interacting with external services, and returning completed results, enabling organizations to automate routine operations such as monitoring, reporting, and data processing.
According to Sber, benchmark testing demonstrates that GigaChat 3.5 Ultra outperforms the company's previous flagship model across programming, mathematics, multi-step reasoning, and Russian-language dialogue tasks while achieving competitive performance against leading open-source AI models. The combination of improved efficiency and reduced hardware requirements is expected to make advanced AI deployment more accessible for businesses, researchers, and software developers.
"GigaChat 3.5 Ultra represents our vision of AI as a practical partner capable of solving real-world tasks rather than simply answering questions," said Anton Frolov, Senior Vice President and Head of GenAI Development at Sber. "By combining our proprietary architecture with extensive research and engineering, we have created a model that delivers stronger performance while using significantly fewer computational resources. We believe this launch will help accelerate innovation across enterprise AI, research, and software development."
The launch reflects Sber's continued investment in artificial intelligence research and open innovation. By making GigaChat 3.5 Ultra available through both its GigaChat AI assistant and as an open-source model, the company aims to support broader adoption of advanced AI technologies while enabling developers and organizations worldwide to build next-generation AI applications.
About Sber
Sber is Russia's largest bank and one of the country's leading technology and financial services companies. In addition to banking and financial solutions, the company develops advanced digital technologies across artificial intelligence, cloud computing, cybersecurity, and enterprise software. Through ongoing investment in research and innovation, Sber continues to expand its AI ecosystem for consumers, businesses, developers, and research organizations.