Ling-3.0-tiny puts 1.3B active parameters on local hardware
Ling-3.0-tiny is a 7.9B-parameter model built to bring reasoning and agent features to local, resource-constrained machines, with just 1.3B parameters active for each token. It uses a sparse mixture-of-experts design: each token selects eight routed experts plus one shared expert, while hybrid reasoning can switch between fast replies and multi-step thinking. The model page reports about 86 to 90 tokens per second on an M4 Pro MacBook and roughly 8.34 GiB peak memory at an 8K context.