← Back to research

Research Brief · 2026-11-09 Adjacent

Joi-Bench-Task: Benchmarking Task Coordination in Full-duplex Voice Agents through Interactive Simulation

The idea, in brief

Five commercial voice agents run background tasks while a simulated user interrupts and changes plans. Most lose track of what they still need to report.

Editorial summary · not the author abstract. Read the linked paper for the complete abstract and claims.
Figure 1: the Joi-Bench-Task framework, in which a simulated user intervenes at set moments during tool execution and spoken result delivery.
Figure 1: the Joi-Bench-Task framework, in which a simulated user intervenes at set moments during tool execution and spoken result delivery. Original figure · paper authors ↗

Release evidence

The paper and arXiv record link no code or data, and a web search found none.

Last checked · 2026-11-15