Inclusion AI, Ant Group's AI research arm, has published the open weights for Ling 3.1 Flash on Hugging Face under a permissive MIT license. It's a sparse Mixture-of-Experts (MoE) architecture with 560 billion total parameters, activating just 25 billion per token (560B-A25B).
That setup gives you the breadth of a massive model with the inference speed of a much smaller one. Its context window stretches to 1 million tokens—enough to swallow entire monolithic codebases or dense financial specs in one go. On benchmarks, Ling 3.1 Flash leads in app interface generation (UI/UX) and math.
Release highlights:
- MoE sparsity: inference compute stays on par with a 25B-parameter model.
- Permissive MIT license: open weights ready for commercial deployment with no hidden strings.
- Ecosystem support: native integration with vLLM, SGLang, Transformers, and TensorRT-LLM.
We've been tracking how Inclusion AI's Ling 3.1 Flash evolved and how earlier versions explored MoE architecture and a 256K-token context. Now the model has scaled up and is open to everyone.
