Comment by antirez

The model performs very poorly in practice, while in the benchmark it is shown to be DeepSeek V3 level. It's not terrible but it's at another level compared to the models it is very close to (a bit better / a bit worse) in the benchmarks.