AI agents were given an office work exam: Sber released the open Cowork Bench benchmark

Sber has released Cowork Bench, an open Russian-language benchmark assessing AI agents' ability to autonomously handle office work tasks across 475 scenarios, 12 business functions, and 26 corporate systems. The benchmark compares different AI agent software shells on a uniform model and tasks, showing varying completion rates from 62% to 80%.

This Wire brief sits within Fusion42's coverage of AI Agents.