Ling-3.1-flash-base

Researchers and developers can use the checkpoint for continued pretraining, domain adaptation, and post-training research.

Availability

Released as a base checkpoint on Hugging Face

Pricing model

Not established

Deployment

Not established

Privacy

Not established

Licence

MIT

• The base checkpoint is released to support research and community-driven innovation. ^[1]

  We introduce **Ling-3.1-flash-base**, the base checkpoint of [Ling-3.1-flash](https://huggingface.co/inclusionAI/Ling-3.1-flash), released to support research and community-driven innovation.

• The model has 560B total parameters and 25B active parameters per token. ^[1]

  With **560B total parameters** and **25B active parameters per token**, the model provides a foundation for continued pretraining, domain adaptation, and post-training research.

• The checkpoint is intended for continued pretraining, domain adaptation, and post-training research. ^[1]

  With **560B total parameters** and **25B active parameters per token**, the model provides a foundation for continued pretraining, domain adaptation, and post-training research.

• The architecture combines KDA with Gated MLA in a hybrid-linear Mixture-of-Experts design. ^[1]

  Ling-3.1-flash-base shares the same hybrid-linear Mixture-of-Experts (MoE) architecture as Ling-3.1-flash, combining KDA with Gated MLA.

• The released checkpoint is BF16, produced after Warmup-Stable and Merge checkpoint merging. ^[1]

  This release provides the **BF16 base checkpoint** produced after **Warmup-Stable and Merge (WSM)** checkpoint merging.

• The registry lists the model license as MIT. ^[1]

• New models bring live speech translation and voice features Daily brief 2026-10-28