Artificial Analysis launched Time to First Audio, a benchmark measuring how long text-to-speech models take before audible speech begins, filling a gap in its existing TTS quality and speed evaluations. Leading models range from 76 ms to nearly one second, with streaming models at a 376 ms median versus 2.0 seconds for non-streaming ones, making streaming the key divide for voice agent responsiveness.
Modelsslug: simba-3-5
Simba 3.5
Speechify's speech synthesis model emphasizing high throughput and low unit price.
Curated Intelligence
Artificial Analysis (@ArtificialAnlys)T1·Tools
62 pts
Artificial Analysis (@ArtificialAnlys)T1·Models
58 pts
Speechify's Simba 3.5 ranks first in the Artificial Analysis Controlled Voice Arena, statistically level with HeyGen Voice and ahead of Qwen-Audio-3.1-TTS-Plus. It delivers 139.6 characters per second at $6.58 per 1M characters, the cheapest model on the price chart, though it places only fifth on the Provider Voice leaderboard.