Cartesia

Cartesia specializes in developing real-time multimodal AI models powered by state space models (SSMs). The company's SSM architecture offers significant advantages over traditional transformer models by scaling linearly with sequence length and enabling efficient, high-throughput inference. Unlike transformers that process every past token, SSMs update the model's state and discard previous tokens as they stream in, making them ideal for real-time applications.

In May 2024, Cartesia launched Sonic, an ultra-fast text-to-speech model that generates expressive, lifelike speech with less than 90 ms latency to first audio. Sonic operates locally without internet connection and includes features like emotion control, speed adjustment, and prompting capabilities. The platform outperforms existing market solutions in voice quality, stability, and accuracy as validated through blind human preference tests by third-party evaluators like Labelbox. The company built and optimized its own SSM inference stack to serve Sonic with low latency and high throughput at scale.

Cartesia's technology enables developers to create AI applications across various sectors including customer service, healthcare, robotics, gaming, transportation, education, and security. The company developed the widely cited Mamba architecture that demonstrates SSMs can match transformer performance using fewer computational resources.

Key customers and partnerships

Sonic API has been adopted by hundreds of customers ranging from startups to public companies for applications including customer service, debt collection, interview screening, voiceovers, and interactive character voices. The platform has gained particular traction among startups building real-time voice agents in the interactive voice response (IVR) market, which was valued at USD 6 billion as of December 2024.

Cartesia partnered with Crusoe Cloud to support the training of their models on H100 GPU clusters, and has expanded this partnership multiple times. The company's Sonic product is also available through the AWS Marketplace.