Ant Group's Bailing open-sources Ling-3.1 enhanced version, 560B pre-training base released simultaneously

Beating AI News Flash: Ant Ling officially open-sources the Ling-3.1-flash-1019 model weights, and also releases Ling-3.1-flash-base. Both models are now available on Hugging Face and ModelScope under the MIT open-source license.

The 1019 version was first announced on October 22, when only the API was available. The new version continues the mixture-of-experts architecture with 560B total parameters and 25B activated parameters, mainly strengthening reasoning and Agent capabilities through post-training. The Artificial Analysis Intelligence Index rose from 41 in the old version to 44. The AA-Briefcase evaluation for complex office tasks improved from about 1400 to 1524 Elo, the AutomationBench-AA software operation evaluation rose from 62% to 68%, and the CritPt physical reasoning test rose from 18% to 26%.

The simultaneously released Base is a foundation model that has not yet undergone full instruction alignment and post-training, released in BF16 precision. Developers can use it for continued pre-training, industry fine-tuning, or reinforcement learning without having to start from an already aligned chat model.