Sber Launches GigaChat 3.5 Ultra With Coding Upgrades

A New Contender in AI Coding Assistance

Russian tech giant Sber has introduced GigaChat 3.5 Ultra, a new version of its AI assistant that the company says brings meaningfully stronger code-generation abilities, better handling of agent-based tasks, and improved processing of long-form text inputs 1. According to the release, the model was tuned for faster performance while also becoming more resource-efficient, a combination that matters as enterprises increasingly weigh the operating costs of running large models at scale 1. For developers and technical teams, the emphasis on coding quality and agentic task execution positions GigaChat 3.5 Ultra as a direct entrant into the same competitive space occupied by Western coding assistants and code-review tools, even as most of the detailed specifications remain limited to Sber's own claims 1.

The Industry Backdrop: Agents and Personality

Sber's update lands amid a broader industry shift in which AI assistants are being judged not just on raw capability but on how they operate as agents and how they interact with users. That shift is visible in Cognition's decision to acquire Poke, a move explicitly aimed at giving its coding agent Devin a more distinctive conversational style, on the theory that personality and interaction design are becoming as competitively important as underlying model quality 2. The same logic underpins Meta's recent moves: the company has rolled out task-automation features that let its AI assistant deliver daily briefings, summarize calendar events, and handle recurring requests like weekly meal plans, initially through the Meta AI app and meta.ai before expanding to more markets and platforms such as WhatsApp 3. Separately, Meta has also described its Muse Spark 1.1 model as a step toward giving its assistant genuine personal-agent capabilities, extending beyond simple query response into proactive task management 5.

Where Established Assistants Fit In

Agentic ambitions are not confined to newer entrants. Google Assistant, long established as a mainstream voice-driven helper, continues to expand its practical utility for everyday tasks such as placing hands-free calls and managing smart-home routines, with guidance emphasizing that thoughtful setup and configuration unlock much of its functionality 46. While Google Assistant's use cases are more consumer-oriented than the developer-focused ambitions of GigaChat 3.5 Ultra or Devin, its continued evolution underscores how deeply agent-like automation is spreading across both consumer and enterprise AI products.

Why It Matters

Taken together, these developments point to a broader convergence: AI assistants across coding, productivity, and consumer voice categories are being pushed toward greater autonomy, longer context handling, and more distinctive user interaction. Sber's push into agentic coding tools mirrors moves by Cognition and Meta, suggesting that competition in AI assistants is increasingly being fought not only on benchmark performance but on usability, personality, and the breadth of tasks an assistant can independently complete.