Build ultra low latency voice AI applications with Together AI and Cartesia Sonic

Today we’re announcing the launch of the Together Audio API, powered by Cartesia Sonic – a state of the art, low-latency and ultra-realistic voice model. Through our partnership with Cartesia, developers can now access the Sonic model directly via the Together API. Along with support for several different voices, Sonic is also multilingual, making it great for any voice use case. This exciting addition expands our platform's capabilities, letting developers build multi-modal applications with chat, image, audio, code, and embeddings, all through a single platform.

Key Highlights

• New Together Audio API powered by  Cartesia Sonic – the SOTA low latency and ultra realistic, multilingual voice model. Sign up and add credits to get started.
• Build enterprise-ready voice applications on the Together Platform with HIPAA and SOC2 compliance.
• Get started with voice applications using our cookbooks – for example learn how to use agentic workflows to create your own NotebookLM style podcast.

{{custom-cta-1}}

Build multi-modal applications on the Together Platform 

The addition of audio capabilities marks a significant milestone in our mission to enable developers to build and orchestrate multi-modal applications. Developers can now integrate multiple AI models—including chat, image, audio, code, and embeddings—using the Together API Platform.

Most voice applications rely on the orchestration of several AI models working together. For example, systems can combine speech-to-text (STT), a large language model (LLM), and text-to-speech (TTS) to deliver conversational experiences. For example, imagine an AI agent rebooking a flight end-to-end: using Retrieval Augmented Generation (RAG), it accesses passenger records, verifies bookings, suggests alternatives, and processes changes—all in a single, seamless interaction. With the Together Platform, developers can orchestrate all these capabilities easily, ensuring the lowest latency, without juggling multiple API providers.

Voice AI use cases

Voice AI is transforming applications across every single industry, be it medical agents to assistant doctors in healthcare, or customer service agents in retail, and 85% of companies anticipate widespread deployment of voice AI systems within the next five years (Source). 

Here are some ways developers can leverage voice capabilities:

AI-powered customer support

Create voice agents that handle customer inquiries naturally and effectively. Combine LLMs for reasoning and contextual understanding, with Sonic’s natural-sounding responses. Frameworks like Daily’s Pipecat make it easy to coordinate these models in real-time applications.

Content creation and media

Automate audio content production with AI. Our python notebook demonstrates how to create podcasts by chaining LLMs with voice generation. This approach is ideal for producing narrated content, translation and dubbing, or summaries of long-form material.

Voice assistants and digital twins

Develop personalized AI assistants with consistent voice and personality. By combining chat and voice capabilities, developers can create assistants that understand context and respond naturally in real-time.

Why Cartesia Sonic 

Cartesia Sonic consistently outperforms other voice models in blind human preference tests, delivering remarkably natural-sounding voices. Here's why developers choose Sonic:

• Ultra low latency: With just 90ms streaming latency - half the time of a human blink - Sonic delivers the fastest end-to-end voice applications (STT → LLM → TTS) available today, with continuous performance gains on the horizon.
• Superior content processing: Masters complex inputs like phone numbers and technical terminology with striking accuracy, while offering diverse voice options across 15 languages.
• Leading Voice AI research: The key to Cartesia's performance breakthroughs is their innovative State Space Model (SSM) architecture, achieving major breakthroughs in voice realism, personalization, and efficiency.

Cartesia hosts its models for inference on Together GPU Clusters, and we’re proud to support them as pioneers of voice AI. We’re excited to deepen this partnership by becoming an official platform providing direct access to Sonic through our API.

Getting started with building voice applications

Ready to build with voice? Here's how to get started:

• Sign up and get your API key at together.ai
• Add credits and try out Cartesia Sonic via our audio playground, or Audio API
• Check out our documentation
• Try our example notebook and create your own notebookLM style podcast 

Join our developer community on Discord to share what you're building — we're excited to see your multi-modal applications come to life.