Ling 3.0 Tiny activates 1.3B parameters

Ant Group’s Ling 3.0 Tiny activates about 1.3 billion of 7.9 billion parameters per token and offers a 256K-token context window with up to 32K output tokens. Vercel’s launch notice says the mixture-of-experts model supports native function calling and prompt caching, and is free through August 14 before its gateway identifier loses the -free suffix. For teams weighing the infrastructure burden behind large open-weight mixture-of-experts models, this is a low-cost reminder to benchmark task success and latency rather than total parameter count: only 16.5% of Ling’s weights are active on each token.