Comment by CSMastermind

DeepSWE is the benchmark you want to actually look out for. Only one that aligns with actual user reported results from trying the models.