[yorwba]

The benchmark scores in the new announcement are significantly higher than for the preview model.

[samus]

That's good news, I was highly impressed already by what that model could do, even under heavy quantization.