Fact · Voice and AI
85percent task completion
On the τ-Voice benchmark, GPT-5 with reasoning completed 85 percent of tasks, against 31 to 51 percent for voice agents on clean audio.
Checked against its source on October 17, 2026
In the source's own words
“while GPT-5 (reasoning) achieves 85%, voice agents reach only 31--51% under clean conditions”
τ-Voice: Benchmarking Full-Duplex Voice Agents on Real-World Domains, Soham Ray et al., arXiv, 2026, Abstract.
Open the original- What it measures
- The benchmark runs tasks with policies and tools, and compares voice and text on the same tasks.
- What it does not show
- A preprint with 278 tasks in its own simulator. Scores depend on the benchmark design and are not a ranking of products.