Sber presents "GigaAgent": an autonomous AI agent with self-development and continuous operation capabilities.
The AI agent market is entering a new phase: simple conversational chatbots are being replaced by full-fledged digital partners capable of planning and executing complex multi-step tasks. In this race, Sber is making an ambitious move by launching "GigaAgent" — a universal autonomous assistant for everyday and work tasks. This launch comes amid intensifying competition in the segment, right after similar announcements from Yandex.
Assistant for routine tasks
"GigaAgent" is positioned as Russia's answer to foreign solutions like OpenClaw, Hermes Agent, and NemoClaw. Unlike traditional chats, the new agent does not require control at every step: it manages calendars, works with documents, handles correspondence, searches for information, and prepares reports with presentations. The key difference is its ability to accumulate context and adapt to the user's individual work style, rather than simply generating responses to queries.
The developers endow "GigaAgent" with three key properties that radically distinguish it from its predecessors:
- Autonomy: the agent operates around the clock in a proactive mode. It honestly reports a lack of data and can raise reasoned objections if it detects an error in reasoning.
- Self-improvement: the system is capable of improving its own work by reading and rewriting its source code. Each change is recorded as a commit (checkpoint) in a built-in repository, ensuring transparency and the ability to roll back to previous versions.
- Continuity: most AI agents operate within a single session and lose accumulated experience. "GigaAgent" remembers past tasks, decisions made, and interaction history, continuing work after a restart with the collected context taken into account.
Technology and pricing
The development is based on the self-improving agent project Ouroboros from the research team at the AIRI Institute. The results are impressive: on the Terminal Bench 2.1 benchmark for Anthropic's Opus 5 model, the agent's quality reaches 86.97%, placing it on par with global leaders.
The service can be launched through the Agents Space environment from Cloud.ru. New users receive a starting grant of 4,000 rubles for free testing of the platform's capabilities. Further usage is billed at rates for language models and infrastructure, with the final amount depending on the volume and frequency of use.
The scale of investment is striking: according to expert estimates, creating the agent — orchestration, connectors, interface, and quality assessment system — required 1 to 3 billion rubles and a team of about 100 people over a year and a half. Sber does not disclose exact figures, but for a company of this scale, it is a deliberate bet on leadership in technological development.
Analytical commentary: Claims about "self-improvement" and "autonomy" sound bold, but the key question is what measurable business metrics they stand behind. While "GigaAgent" demonstrates impressive benchmarks, its real value will be determined by its ability to reduce operational costs and create new revenue streams. In conditions where every request to the agent has a cost, and errors require oversight, success will depend not on marketing promises, but on clear mechanisms for evaluating effectiveness and managing risks. This is Sber's systemic bet on AI, and it deserves particularly close attention.