Ant Group open-sources Ling-3.1-flash-1019 and its base model

Ant Group's Ling team has released Ling-3.1-flash-1019 and the underlying pretrained Ling-3.1-flash-base on Hugging Face under the MIT license. The hybrid linear mixture-of-experts model keeps about 560 billion total parameters, roughly 25 billion active per token and a one-million-token context, with gains coming from extra post-training for reasoning, tool use and multi-step execution.

• Scores 44 on Artificial Analysis Intelligence Index v4.3.2, up three points
• AutomationBench-AA rose from 62% to 68%, CritPt from 18% to 26%
• Ant says the base model beat DeepSeek-V4-Pro-Base on 32 of 47 benchmarks
• BF16 weights support continued pretraining; serving via SGLang and vLLM on 8-GPU nodes