Pluralis Research

Pluralis Research is an AI development company focused on decentralized approaches to training large-scale models. Its core innovation, Protocol Learning, enables multiple participants to collaboratively train foundation models without any single party obtaining a full copy of the model. The company operates globally.

Protocol Learning is built on model-parallel training techniques designed to distribute computational workloads across geographically dispersed devices. The approach addresses concerns around data privacy, model ownership, and centralised control in the development of large AI systems. Pluralis Research has successfully demonstrated this method by training an 8B-parameter LLaMA model to performance on par with centralised training, using devices in four different locations.

The company's work has been published at NeurIPS 2025 and ICML 2025, two leading venues in machine learning research. Pluralis Research raised $7.6 million in seed funding to advance its technology.