CAMB.AI Unveils MARS8, a Multi-Architecture Text-to-Speech Platform Built for Enterprise Production

TL;DR

• CAMB.AI, the AI voice infrastructure company founded by former Apple and Carnegie Mellon researchers, on Tuesday announced the launch of MARS8, a next-generation text-to-speech (TTS) platform.

CAMB.AI Unveils MARS8: The First Family of TTS Architectures, Ending the Era of One-Size-Fits-All Voice AI

CAMB.AI, the AI voice infrastructure company founded by former Apple and Carnegie Mellon researchers, on Tuesday announced the launch of MARS8, a next-generation text-to-speech (TTS) platform designed as a family of specialized architectures rather than a single, one-size-fits-all model.

The release reflects a key insight gained from years of deploying voice AI at scale: no single TTS system can optimally meet the competing demands of speed, quality, cost, and accuracy across all use cases.

Since 2023, CAMB.AI’s MARS engine has powered AI voice-enabled experiences for more than 200 million users worldwide through partners including NASCAR, Broadcom, the Australian Open, Eurovision Sport, IMAX, and Comcast NBCUniversal. After shipping voice AI into high-stakes production environments across more than 150 languages, the company concluded that the industry’s reliance on generic APIs was limiting real-world performance.

“The market forces developers to choose between speed, quality, accuracy, and cost — and we realized that was a false choice,” said Akshat Prakash, CTO of CAMB.AI. “A live voice assistant needs sub-150 millisecond latency, a movie dubbing pipeline needs director-level emotional control, and an automotive system has strict memory constraints. You can’t solve all of that with one generic model.”

A Family of Models for Real-World Constraints

MARS8 replaces the traditional “black box” TTS API with four purpose-built architectures, each optimized for a distinct production scenario:

• MARS-Flash: An ultra-low-latency model delivering sub-150ms time-to-first-byte, designed for real-time voice agents and call centers, and optimized for Blackwell GPUs as well as L4 and L40S hardware.

• MARS-Pro: A balanced model built for expressive dubbing and digital media, combining high fidelity with production-grade speed.

• MARS-Instruct: A high-end system offering director-level control for film and premium content, allowing independent tuning of speaker identity and prosody.

• MARS-Nano: A lightweight, 50-million-parameter model designed for on-device applications where compute and memory are constrained.

Shifting the Economics of Voice AI

Alongside the technical launch, CAMB.AI is introducing a “compute-first” business model that challenges the industry’s dominant pay-per-character pricing approach. According to the company, token-based pricing scales linearly with usage and erodes margins at enterprise scale.

With MARS8, customers can deploy models on their own infrastructure — including AWS Bedrock, Google Cloud Vertex AI, and specialized GPU platforms such as Modal and Baseten — or via on-device SDKs. At launch, MARS8 is available across more than 25 compute platforms, a first for the voice AI sector.

“Compute-based pricing flatlines costs even as usage spikes, saving enterprises up to 90%,” CAMB.AI said, while also allowing customers to control latency and keep data within regional compliance boundaries.

Partner Ecosystem and Enterprise Focus

MARS8 debuts with an ecosystem of partners spanning cloud infrastructure providers and voice agent platforms, enabling rapid deployment in applications ranging from contact centers to media production pipelines.

CAMB.AI positions MARS8 as a turning point for production-grade voice AI, arguing that the era of forcing diverse use cases through a single generic model is coming to an end.

Whether powering real-time voice assistants, dubbing feature films, or enabling on-device speech systems, MARS8 is designed to meet specific operational constraints without compromise — while making large-scale deployment economically viable.

More details on the MARS8 family and technical architecture are available at camb.ai/mars and in the company’s technical report.

About CAMB.AI

CAMB.AI is a localization-focused AI infrastructure company built to help global enterprises reach audiences across text, audio, and video. With proven deployments in live sports, cinema, and broadcasting, the company aims to enable real-time, language-agnostic access to content for billions of users worldwide.