Artificial Analysis launches Time-to-First-Audio metric: Eleven v4 Turbo hits parity at 258 ms

ArtificialAnlys · x · 2026-11-17

Artificial Analysis introduced Time to First Audio (TTFA), a new benchmark measuring how long TTS models take before a user hears anything — the number that decides whether a voice agent feels responsive. Leading models range from 76 ms to nearly a second.

On the quality-vs-latency Pareto frontier:

The comparison chart spans TTS models from AssemblyAI, Cartesia, Fish Audio, Google, Gradium, Inworld, MiniMax, Mistral and others.

Related event: Artificial Analysis Launches TTS Time-to-First-Audio Benchmark(4 posts)→

Original post →

More from Multimodal

Multimodal channel →