According to CoinMeta, Ant Birdling ( Inclusion AI ) has released the ling-3.0-tiny small model, which is a compact MOE model designed for local deployment and agent scenarios. It boasts a total of 7.9 billion parameters, with only 1.3 billion parameters activated in each token. This model supports both thinking and quick-response modes and can also invoke tools to execute tasks. Although only 1.3 billion parameters are activated at a time, it has already outperformed larger Qwen3.5-9B models on certain agent tasks. In official tests, ling-3.0-tiny achieved higher scores in two general agent evaluations: gdpval, v2-aa, and tau3-banking-aa. Currently, ling-3.0-tiny has been launched on OpenRouter and Vercel AI Gateway, and users can try it out for free. The model weights will also be made open-source in the future.