Ling Mini 2.0
InclusionAI · 2025-09-10 · 16.3B parameters
Ling Mini 2.0 is InclusionAI's (Ant Group's) compact MoE language model from the Ling 2.0 family, with approximately 16 billion total parameters and 1.4 billion active per token. Trained on more than 20 trillion high-quality tokens and refined through multi-stage supervised fine-tuning and reinforcement learning, it delivers strong complex reasoning and instruction-following capabilities relative to its small active parameter count. It is designed for cost-efficient deployment where lower computational overhead is a priority.