Rime launches Prism, a speech-to-speech model for voice agents

Rime launched Prism, a speech-to-speech model for AI voice agents. The figures below are Rime's own claims, not independently measured.

• Speed: a response time under 300 ms, which Rime says is 4x faster than OpenAI's GPT-Live.
• Intelligence: Rime says it outperforms every other speech-to-speech model on intelligence, without reasoning, and calls it the first speech-to-speech model customizable through reinforcement learning.
• Availability: developers can use it through Rime's API. Rime offers one month free to the first 100 people who sign up.

TestingCatalog's write-up of the launch adds that Prism writes each reply as text before speaking it, so companies can edit the transcript and block things an agent should not say. While a lookup runs it says a short line such as "Let me check that." It supports only English, with four voices, and works with LiveKit and Pipecat.

Rime says it is a team of 35 with $35M in funding.