AssemblyAI-TTS-1 lands on OpenRouter for voice agents
- Source
- x.com
- Date
AssemblyAI has released its first text-to-speech model, AssemblyAI-TTS-1, aimed at voice agents. OpenRouter now routes to it, which puts it behind the same API that teams already use for language models, with more than 4,000 voices and a choice of US or EU serving. Artificial Analysis ranks it #12 in its Provider Voice Arena and #10 in its Controlled Voice arena, both blind preference tests. It also scores 81.7% on pronunciation robustness and measures 162 characters per second at $15 per million characters. OpenRouter cites a separate Cekura benchmark that rates the model well on speed, accuracy and naturalness, a claim that has not been independently checked. The Artificial Analysis figures give builders a common scale for comparing it with the higher-priced ElevenLabs and Cartesia voices already in use.
AssemblyAI-TTS-1 is now live on OpenRouter! @AssemblyAI's first text-to-speech model, built for voice agents. Ranked in the best quadrant for speed and accuracy, and highest naturalness for models at $15/1M characters or less, per @cekuraAi's benchmark. 4,000+ voices, served from US or EU.
Context
OpenRouter made AssemblyAI-TTS-1 available, the first text-to-speech model from AssemblyAI. Text-to-speech turns written text into spoken audio, and OpenRouter says this model was built for voice agents. It offers more than 4,000 voices and can be served from the US or the EU.
OpenRouter's quality claim comes from a by Cekura: the model ranked in the best quadrant for speed and accuracy, and had the highest naturalness among models priced at $15 per million characters or less. The Cekura results were not available for this explanation, so both rankings are OpenRouter's summary rather than checked figures. The price condition also narrows the naturalness claim, since it compares the model only with models that cost the same or less, not with the whole field.
A separate by Artificial Analysis, covered on Intel, gives a more mixed picture of the same model. It ranked AssemblyAI-TTS-1 12th in its Provider Voice Arena with an Elo of 1,227 and 16th on pronunciation robustness at 81.7%, measured generation at 162 characters per second, and listed its price at $15 per million characters. That price puts the model at the top edge of the band in OpenRouter's naturalness claim. OpenRouter's post does not state what the model costs on OpenRouter.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
Checking sign-in…
Loading comments…