GigaChat 3.5 Ultra Can Write Code, Solve Complex Tasks – and It’s Now More Efficient Than Before
Russia's Sber has introduced its new flagship model, GigaChat 3.5 Ultra. According to its developers, the model can solve mathematical problems, perform financial calculations, analyze and evaluate regulatory documents, and both write and execute code.
Sber's latest model is now roughly half the size of its predecessor and requires significantly fewer computing resources. According to the development team, professionals across a wide range of industries will be able to speed up their work by delegating more routine and analytical tasks to artificial intelligence.
The Open Version Is Available Free of Charge
GigaChat 3.5 Ultra is Sber's proprietary large language model. It is available through the GigaChat AI assistant for both personal and professional use. Meanwhile, developers can use its open version free of charge when building applications, AI services, and autonomous AI agents.
The model's defining innovation is its Russian-developed linear-attention architecture. Unlike conventional transformer models, where each new token is compared against all preceding text, this architecture gradually accumulates contextual information as processing continues. That design makes GigaChat 3.5 Ultra nearly twice as compact as the previous generation, reduces computational requirements, and generates long-form text roughly four times faster.
During development, the engineering team conducted approximately 1,500 experiments. The updated model demonstrates stronger performance in code generation and validation, mathematical reasoning, and financial calculations. Because its responses have become more accurate, analysts and software developers can integrate the model more confidently into production workflows. GigaChat 3.5 Ultra also analyzes technical regulations, reports, contracts, and other complex documents while preserving their meaning. The model can accumulate knowledge over time and understand the context in which that knowledge was originally introduced.
An AI Tool for Everyday Professional Work
Building AI agents has also become simpler. Once users define a task, GigaChat 3.5 Ultra can retrieve the required information, write and execute code, call external services, and return the final result. That workflow allows organizations to improve productivity by offloading routine tasks to an AI assistant. According to Sber, the model now approaches the performance of leading open-weight models on several benchmarks, including DeepSeek 3.2. Company representatives said the latest improvements have transformed the system into an AI tool designed for real-world workloads rather than conversational interactions alone. Over time, the neural network is expected to evolve into a full-fledged digital partner capable of reasoning within the logic of a specific business process instead of simply answering questions.
Lower Costs for Businesses
The updated GigaChat 3.5 Ultra is more than another chatbot upgrade. It represents continued development of Russia's foundational generative AI technology. Individual users gain free access to an AI assistant for programming, document processing, and complex calculations, while businesses can reduce the cost of document processing, analytical work, and software development. The updated model can also be deployed in educational platforms, banking services, and government digital platforms.
For Russia's software industry, what matters most is that the model uses its own architecture and is available openly, reducing developers' dependence on foreign language models and overseas cloud infrastructure. Continued development of Russian large language models strengthens the country's technological sovereignty while supporting broader adoption of artificial intelligence across government, industry, education, and other sectors.
Looking ahead, Sber's software platform could become a competitive export offering for CIS countries and other markets where Russian-language digital solutions are in demand. Additional opportunities may emerge through Russia's cooperation with BRICS countries, where artificial intelligence and industrial innovation are viewed as promising areas for long-term technological collaboration.
