Ant Group open-sources Ling-3.1-flash under MIT: 560B total, 25B active, 1M context

Ant Group's inclusionAI released the Ling-3.1-flash weights on Hugging Face under the MIT license. The hybrid MoE model has 560B total parameters with 25B active per token and a 1M-token context window; its free API trial ended on October 14.

• 560B total parameters, 25B active per token
• Context window up to 1 million tokens
• 63 KDA linear-attention layers and 9 Gated MLA layers in a 7:1 ratio
• Recommended SGLang setup: eight H200 GPUs or four GB300 GPUs