Compact vision-language MoE model from Baidu — 28B total, 3B active.
Add a second model to compare.
This model hasn't been benchmarked yet.