Ling 3.0 Flash, the smartest open model yet! Released by Ant Group
Ling 3.0 Flash — Top Open Model (<124B params). AAI: 38. Hallucination Rate: 97% -> 44%. MIT License.
InclusionAI, a subsidiary of Ant Group (financial company owned by Alibaba), published their latest model, Ling 3.0 Flash. It’s an MoE (Mixture of Experts) model with less than 124 billion parameters (only part of them activated at the same time), open source under the MIT license, and the weights are available on Hugging Face and DeepInfra.
Benchmarks: Top of Its Class
It achieves an AAI score of 38, making it the highest scoring open model of its size. The hallucination rate decreased from 97% to 44%. It’s also cheaper per token than any other model of similar quality.
DeepSeek V4 Flash has an AAI score of 52, so while Ling 3.0 Flash is still behind in raw performance, it’s much more affordable. We can see that open Chinese models are getting closer to the state-of-the-art by optimizing their architectures instead of just scaling up.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Chief Executive Officer, digital health
Articles · Latest articles