San Francisco-based Rime has introduced Prism, a speech-to-speech model designed for enterprise applications. The model processes audio input directly into audio output, bypassing the traditional multi-step pipeline of transcription and reasoning. A key feature includes a verification checkpoint that generates text before the final audio response, allowing businesses to audit the model's output.
Prism supports fine-tuning on proprietary client data and can be deployed in private clouds or on-premises environments. The model achieves a response time of approximately 500 to 570 milliseconds. In customer service benchmarks, Prism scored 56.6 on the τ-Voice test, which the company claims outperforms real-time models from OpenAI and Google.
During a trial with Revmo, Prism successfully completed 94% of order placements, compared to 57% for an OpenAI model. The system also achieved 100% order confirmation rates and reduced hallucinations to 85% of calls. Rime currently manages nearly 100 million phone calls monthly for clients in sectors including healthcare, finance, airlines, and food service.
The company was founded by Lily Clifford, Brooke Larson, and Ares Geovanis. Rime recently announced a total fundraising amount of $35 million. Publicly disclosed records indicate a $24 million Series A round in July 2026 and an earlier $5.5 million seed round, with total funding figures cited between $29.5 million and $35 million.
