What Zyphra does
Zyphra is an open-source, open-science AI company headquartered in Palo Alto, California, founded in 2020. It focuses on novel neural network architectures, long-term memory, and continual learning, with the goal of building a superintelligence lab. Zyphra develops efficient multimodal foundation models spanning language, vision, and audio, and is building Maia, a general-purpose superagent designed to boost productivity for enterprise knowledge workers.
Key capabilities
Zyphra is known for state-space-model (SSM) hybrid architectures. Its open Zamba models combine Mamba blocks with a global shared attention layer and reportedly outperform comparable open models like LLaMA 2 7B and OLMo-7B while using less than half the training data. The company also released the Zonos text-to-speech and voice-cloning model and the open Zyda datasets. In 2025 Zyphra raised a $100M Series A at a $1B post-money valuation led by Jaan Tallinn, with participation from AMD, IBM, and others, and is partnering with IBM and AMD on next-generation AI infrastructure.
Who it's for
Zyphra serves researchers and enterprises interested in efficient open models, on-device AI, voice cloning, and an emerging multimodal superagent for knowledge work.