0G Labs Begins Public Retrain of 107B Parameter AI Model
While the market celebrated Bittensor’s 72B model this week, 0G Labs trained a 107B parameter model eight months earlier using standard 1 Gbps internet.
J. Huang (NVIDIA CEO) just validated decentralized AI on the All-In Podcast, but 0G had already proven frontier-scale training is possible months earlier. One builds a model, the other builds infrastructure.
The team has now started public retraining with full transparency and an open-source commitment. At 48% larger than Bittensor’s Covenant-72B, it is the largest decentralized AI model on record.
DiLoCoX-107B runs on 0G’s full-stack blockchain for AI agents, including an EVM-compatible L1 chain, decentralized compute, distributed storage, and a data availability layer that is 50,000x faster and 100x cheaper than Ethereum's DA.