AssemblyAI-TTS-1 debuts: top-12 voice arena rank at $15 per 1M characters

AssemblyAI has released its first text-to-speech model, AssemblyAI-TTS-1, and in independent testing by Artificial Analysis it lands just inside the top dozen rather than at the top, at a price lower than most of the models ranked above it. All of the figures here come from Artificial Analysis's post. AssemblyAI's own announcement was not available, so the model's features beyond these measurements are not covered.

Artificial Analysis runs two listening arenas and scores each model with an Elo rating, a number computed from how its clips fare in pairwise comparisons. In the Provider Voice arena, AssemblyAI-TTS-1 scored 1,227 over 1,626 appearances and ranked #12, one point behind Breeze TTS 2 (1,228) and ahead of Mistral's Voxtral 3 TTS (1,221) and Inworld's Realtime TTS-2 Flash (1,218). It also ranked #12 in both US and UK English. In the Controlled Voice arena, where the models are compared using the same 8 cloned voices, it scored 1,110 over 1,674 appearances and ranked #10, behind Cartesia's Sonic 3.5 (1,118) and ahead of Speechify's Simba 3.2 (1,086). A third test, Pronunciation Robustness, put it at 81.7% across 2,115 rated clips, #16 overall, behind HeyGen Voice (83.2%) and ahead of Inworld Realtime TTS-2 (80.7%). It did better on the sub-category for preserving exact sequences, ranking #6 at 81.5%.

Price is where the model stands out. AssemblyAI-TTS-1 costs $15 per million characters. Of the eleven models ranked above it in the Provider Voice arena, only two are cheaper: Gemini 3.8 Flash-Lite TTS at $11 and Simba 3.2 at $6.6. For comparison, Artificial Analysis lists Eleven v4 at $80 and Sonic 3.6 at $49. It is also quick, generating 162 characters per second against 148.6 for Sonic 3.6 and 112.3 for Voxtral 3 TTS. On these results, AssemblyAI's entry is a mid-pack voice at a low price, not a new quality leader.