Ant Group unveils 124B open-weight model Ling 3.0 Flash

Ant Group unveils Ling 3.0 Flash

Ant Group has released Ling 3.0 Flash, an open-weight language model. Benchmark evaluator Artificial Analysis reported that the model has a total of 124B parameters and scored 38 on its composite metric, the Artificial Analysis Intelligence Index.

Improvement over the previous generation

Artificial Analysis assessed that Ling 3.0 showed a "marked improvement" over its predecessor. However, the specific score or parameter scale of the previous-generation model used for comparison was not disclosed in the post.

Position in terms of parameter efficiency

What stands out is not the absolute score but the parameter efficiency. Artificial Analysis explained that Ling 3.0 Flash sits on the Pareto frontier along the axis of Total Parameters relative to Intelligence. This means no other model achieves the same level of intelligence score with fewer total parameters. That said, details such as active parameter count, architecture, licensing terms, and training data were not confirmed in this announcement.

Item	Confirmed value
Model	Ling 3.0 Flash
Total parameters	124B
Intelligence Index	38
Weight release	Open-weight
Pareto frontier	Positioned on total parameters vs. intelligence

An extension of Chinese open-weight competition

Ant Group, a fintech company that split off from the Alibaba group, has recently been releasing its own model lineup as open weights. The Ling series is reportedly part of that trend. This announcement can be seen as another example of ongoing competition within the open-weight camp along the axis of performance per parameter.

For more on recent trends in the open-weight model ecosystem, see METAL LAB's model category coverage.

What remains to be confirmed

The information disclosed so far is based on a single post from Artificial Analysis. Further confirmation is needed on detailed benchmark scores by category, context length, support for reasoning mode, and deployment channels.