Prism
Rime launched Prism, which it calls the first fine-tunable speech-to-speech model built for enterprises, with a verifiable text checkpoint before it speaks.
Prism takes in audio and produces audio without separate transcription, reasoning and voice systems. Before it speaks it generates a text output that can be checked, so companies get a record of what it intends to say. It can be fine-tuned on a client's own data and deployed in a private cloud or fully on-premises.
- Date
- Wednesday 28 October 2026
- Lab
- Rime
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| τ-Voice benchmark score | 56.6 Rime says this is ahead of real-time models from OpenAI and Google | company |
| Response latency to first audio input | about 500 to 570 ms | company |
| Order placement completion after consent (Revmo trial) | 94% An OpenAI model in the same comparison managed 57% | company |
| Hallucination-free calls (Revmo trial) | 85% | company |
Access type is inferred from enterprise private cloud and on-premises deployment; the page does not state API terms. The $35 million figure comes from Rime's X post, and the article notes that public rounds add up to less.
Sources
This record was checked against its sources on 29 October 2026. How we check