Ling-3.1-flash-1019 weights are out, and they released the base model too
retained text
Well, “coming soon” turned out to be five days. I posted [the earlier announcement](https://www.reddit.com/r/LocalLLaMA/comments/1xcftoo/ling31flash_is_getting_a_1019_update_with_open/) last week. The [1019 weights are now on Hugging Face](https://huggingface.co/inclusionAI/Ling-3.1-flash-1019). Still 560B total, 25B active, with a 1M context window. This is the update they reported at 44 on the AA Intelligence Index v4.3.2, up from 41. AutomationBench-AA went from 62% to 68%, according to the release. They also put out [Ling-3.1-flash-base](https://huggingface.co/inclusionAI/Ling-3.1-flash-base) in BF16 for further training. Glad they included that. It gives people a starting point for their own post-training instead of having to start with the finished chat model. One small gotcha: the SGLang example in the 1019 README currently still points to `inclusionAI/Ling-3.1-flash`. Change the model path to `inclusionAI/Ling-3.1-flash-1019` if you're copying that command for the new checkpoint.
document fetched 2026-10-28 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1xgrr8n/ling31flash1019_weights_are_out_and_they_released