Orion-16B passes 100B tokens, largest LLM pretrained with decentralized compute

const_reborn · x · 2026-08-15

Orion-16B has surpassed 100B training tokens and continues to improve. It is the largest LLM pretrained using DPP, reaching 10^22 FLOPS with heterogeneous commodity GPUs from permissionless global providers. The team plans to add new GPU types and scale to 256 GPUs next week.

Original post →

More from Infra

Infra channel →