Ling 3.0 Tiny MoE model released with 7.9B total, 1.3B active params

Ling 3.0 Tiny is a 7.9B-parameter MoE with 1.3B active per token, 256K context, and 32K max output. Community benchmarks report it rivaling Qwen3.5 9B reasoning at ~36 tokens/s on low-end PCs. llama.cpp support landed, and Vercel's AI Gateway offers it free until August 14.