Sber introduces GigaAgent, an autonomous AI agent that works without constant human supervision
Сбер
Anthropic
Sber has introduced GigaAgent, which it calls Russia's first autonomous universal AI agent. The system handles documents, correspondence, research, reports, presentations, and calendar management, while remembering context even after restarts. On the Terminal Bench 2.1 benchmark, GigaAgent achieved 86.97% quality when paired with Anthropic's Opus 5 model.
Sber has unveiled GigaAgent, an autonomous universal AI agent designed for everyday and work tasks, according to the company's press service. The system can work with documents, conduct correspondence, research information, create reports and presentations, and manage a calendar. Developers call GigaAgent the first autonomous universal AI agent in Russia. Key features highlighted by Sber include autonomy, self-development, and continuity. The agent retains a history of previous tasks, decisions, and interactions with the user, allowing it to resume work after a restart with accumulated context. GigaAgent can also improve itself by reading and rewriting its own source code, with changes tracked in a built-in Git repository, which helps recovery after critical errors. Unlike ordinary chatbots, it independently breaks tasks into subtasks, asks clarifying questions, and performs multiple actions simultaneously, and can write code if needed. The technology is built on the self-developing AI agent project Ouroboros from the Russian research team at AIRI Institute, with Sber specialists now contributing to the framework. In tests on Terminal Bench 2.1 for Anthropic's Opus 5 model, GigaAgent achieved an 86.97% quality score. For data storage and processing, GigaAgent uses Russian platforms, including Agents Space.
- Abbreviations
- Git = Git
Source: Rusbase (RB.RU) —
original
