Sesame releases TurnBench, an open benchmark for voice AI turn-taking
realmrfakename · x · 2026-08-29
Sesame released TurnBench, an open-source benchmark for evaluating turn-taking in spoken dialogue: whether voice AI knows when a user's turn is truly over, and whether an "mm-hmm" is agreement rather than an interruption.
- The typical gap between human turns is only 200 ms, shorter than the 600 ms needed to plan a single word — listeners predict turn ends from prosodic and linguistic cues rather than reacting to silence.
- Interruptions, talking over users, and awkward silences remain common failures that instantly break the illusion of a natural voice assistant.
- TurnBench mirrors Sesame's internal evaluation methodology (same taxonomy, tasks, scoring), with a public leaderboard, interactive conversation viewer, and self-serve dev-set scoring, built with partners including Carnegie Mellon University and Mundo AI.
Related event: Sesame Releases TurnBench for Evaluating Voice AI Turn-Taking(2 posts)→
More from Models
- Stefano Ermon teases 'largest and most capable model' landing next week, opens early access — StefanoErmon · 2026-10-16
- Inception teases its biggest model yet after Mercury 2 makes TIME's Best Inventions — StefanoErmon · 2026-10-16
- Diffusion LLM Mercury 2 named to TIME's Best Inventions of 2026 — StefanoErmon · 2026-10-16
- Gemini 4 Argon scores 97.5% on ARC-AGI-3, second lab past 95% milestone — fchollet · 2026-10-16
- New inference engine 'e' hits 10,500 tok/s on a single DGX Station — brandon_galang · 2026-10-16
- NVIDIA's Nemotron 3.5 Super VL hits CoreWeave: 12B-active multimodal MoE at $0.25/M input tokens — wandb · 2026-10-16