Ling 3.0 Tiny
InclusionAI · 7.9B parameters
InclusionAI's compact Ling 3.0 language model, supporting thinking and direct-response modes through enable_thinking. Its checkpoint contains 7,893,389,856 parameters after excluding 2,944 routing-bias buffer elements; the card reports approximately 1.3B active parameters per token. The release card supports 256K context, while the default configuration uses 131,072 positions.