• Model specifications: Includes Efficient 2 Billion (E2B), Efficient 4 Billion (E4B), 26 Billion Mixture of Experts (MoE), and 31 Billion Dense (31B) versions, adapted for different hardware from mobile devices to servers.
  • Core capabilities: Designed for advanced reasoning, intelligent workflows, and offline edge-side operation, supporting over 140 languages and code generation.
  • Deployment scenarios: Small models can run with low latency on devices like Android phones and Raspberry Pi, while large models support efficient inference on workstations.