1.6T MoE with ~48B active parameters per token. Trained on 50,000 domestic accelerator cards consuming 30T+ tokens. Native 1M context. Strengthens Meituan position in agentic workflows and software engineering within consumer ecosystem.
1.6T MoE with ~48B active parameters per token. Trained on 50,000 domestic accelerator cards consuming 30T+ tokens. Native 1M context. Strengthens Meituan position in agentic workflows and software engineering within consumer ecosystem.