Most AI voice systems look the same on the surface: a single Text-to-Speech model hidden behind an API. Easy to use, hard to customize, and nearly impossible to scale efficiently.
CAMB.AI took a different approach. Built by Apple, Carnegie Mellon AI researchers, MARS8 is the first TTS family designed for production reality -not lab demos. Instead of forcing one-size-fits-all tradeoffs, MARS8 offers 4 architectures optimized for real-world use cases:
- MARS-Flash: Sub-150ms TTFB for real-time agents.
- MARS-Pro: Balanced speed and fidelity for dubbing and media.
- MARS-Instruct: Fine-grained, director-level control for film production.
- MARS-Nano: Ultra-efficient, 50M parameter model for on-device performance.
MARS8 also changes the economics of Voice AI, moving from a “pay-per-character” pricing model to a compute-first approach, where developers can deploy on their own infrastructure across AWS, Google Cloud, Modal, Baseten & 25 more GPU compute platforms. Start building now!
