0G Labs Quietly Trained Largest Decentralized AI Model in 2025, 48% Bigger Than Bittensor's Recent Breakthrough

Event summary

  • 0G Labs trained a 107 billion parameter decentralized AI model in July 2025, 48% larger than Bittensor's recent 72B parameter model.
  • The model, DiLoCoX-107B, was trained using technology developed with China Mobile, achieving 357x greater communication efficiency than standard methods.
  • 0G Labs claims a 95% cost reduction compared to centralized approaches used by OpenAI, Google, and Meta.
  • The company is now publicly retraining DiLoCoX-107B with a commitment to open-source release and full transparency.
  • 0G Labs has raised $40 million in seed funding and has a $250 million token commitment from investors.

The big picture

0G Labs' achievement underscores the growing viability of decentralized AI training, challenging the dominance of centralized approaches by major tech players. The company's commitment to open-source and transparency sets a new standard for verifiable AI development, potentially reshaping the industry's governance and market dynamics. With significant investor backing, 0G Labs is positioned to drive further innovation in decentralized AI infrastructure.

What we're watching

Decentralized AI
How 0G Labs' achievement will impact the adoption of decentralized AI training methods and challenge centralized approaches.
Cost Efficiency
Whether 0G Labs can sustain its claimed 95% cost reduction as the scale of decentralized AI models increases.
Open-Source Commitment
The pace at which other AI infrastructure projects will follow 0G Labs' lead in open-sourcing large-scale models.