opal-v2 is another accuracy-prediction model.
It's an improvement over the previous opal, which used a heavier algorithm, thus slower and larger model.
The clearest difference you can see is the fact that opal v2 offers a more explainable approach to skill/rating estimation while opal v1, because of its internal complexity, requires a lot of black-box testing to achieve this.
The details of implementation of this model is mostly in my blog post: opal v2: Explainable Map and Player Embedding Optimation with Guided Constraints
As usual, opal doesn't provide recommendations, however, you can try out AlphaOsu!.