The main idea here is: "birds of a feather flock together" That is, words that appear near each other inform the "function" of a word.
More importantly, I think of the techniques you mentioned as "methods," rather than "models." The reason why is it seems possible to violate the definition of what a distributional semantic model is without appropriate data preprocessing.
SVD for example, is typically a dimensionality reduction technique, or the pre-cursor to a clustering technique depending on feature engineering and usage of the methodology. In this case, if you were counting the co-occurrence of words across documents -- rows = documents, cols = words, cells = # of times that word appeared in that particular document -- and then you ran SVD on that, you might have the pre-cursor to a "model" that may eventually be something you can call distributional semantic.
Another example could be Word2Vec, where one typically uses a neural network to train a shallow layer, and extract the weights. Word2Vec can be trained via Skip-gram, or continuous bag of words. Since the "meaning" of a word is derived from the co-occurrence and/or proximity to neighboring words, it may be considered a distributional semantic model. FastText is probably even more so, since it explicitly uses the distribution of words in documents to perform a similar vector operation.
Latent Dirichlet Allocation may be another example. Perhaps even Naive Bayes, if used in the appropriate manner.
So ultimately, the answer is, it depends on data pre-processing/feature-engineering and usage, rather than the technique.