Alas, the article is behind a paywall, so its hard to find out what the full context is, but there is at least one thing I can say.
Usually when a statistical physicist says that they have "solved" a model, it means that they have obtained an (asymptotically exact) formula for the full partition function or free energy as a function of the parameters, including temperature. At any finite temperature one will typically have that the full partition function includes contributions from all possible assignments on the spins (corresponding to all possible true/false assignment in the SAT problem), so that actually computing the partition function looks more like solving a weighted #SAT problem. It is this full problem that was solved by Ising in the general 1D case and by Onsager in a special class of 2D cases. However, if one consider the limit in which the temperature goes to zero the contributions of the terms in partition function with energy larger than the ground state will become negligible. In this limiting case it may turn out the full partition function becomes a number that is a function of the smallest possible number of clauses violated in the formula. When no clauses are violated one is in the ground state and the free energy is zero, which give the mapping to SAT.
If this were the end of the story then it would seem, as you have pointed out, that there is nothing new about the result. However, it needs to be emphasized that this is an approximation, and one that you would only expect to be reasonable at low temperatures. It is typical that, in addition to the ground state, one needs to account for states near to the ground state that give some contribution. There are different techniques (in the wein of Taylor expansions) to handle this in different situation, but AFAIK none of them work in every situation, and certainly aren't universal in the sense suggested in the article. So if there is a proof that all Ising models in the low energy regime are basically the same and can be solved with general these SAT-like techniques this is a new result.
It seems that I wasn't wrong about the context (indeed they focus on solving the full partition function, as opposed to just the ground state energy), but there is some additional information in the paper that is worth sharing.
The first is that their construction relies on a method where (1) the spins used to simulate a target Ising model are a subset of those taken in their universal model based on SAT and (2) that to make a correspondence between the energy levels in the universal mode and the target model one must throw away terms in the universal models partition function with energy larger than some Δ, so that the spectrum of the target model is a subset of the spectrum of the universal model. I think (1) is a neat trick, although maybe not too surprising if you are familiar with, e.g. hidden spin layers in neural network models. With respect to (2) they claim that this allows one to approximate the target model partition function with an error term of size O(exp(-Δ)). I would assume that this means that there can;t be too many states in the full model with with energy greater than Δ, since if they grew exponentially in number this could counteract the exponential damping in contribution in the partition function. Maybe someone who actually read the paper from start to finish can comment?
Usually when a statistical physicist says that they have "solved" a model, it means that they have obtained an (asymptotically exact) formula for the full partition function or free energy as a function of the parameters, including temperature. At any finite temperature one will typically have that the full partition function includes contributions from all possible assignments on the spins (corresponding to all possible true/false assignment in the SAT problem), so that actually computing the partition function looks more like solving a weighted #SAT problem. It is this full problem that was solved by Ising in the general 1D case and by Onsager in a special class of 2D cases. However, if one consider the limit in which the temperature goes to zero the contributions of the terms in partition function with energy larger than the ground state will become negligible. In this limiting case it may turn out the full partition function becomes a number that is a function of the smallest possible number of clauses violated in the formula. When no clauses are violated one is in the ground state and the free energy is zero, which give the mapping to SAT.
If this were the end of the story then it would seem, as you have pointed out, that there is nothing new about the result. However, it needs to be emphasized that this is an approximation, and one that you would only expect to be reasonable at low temperatures. It is typical that, in addition to the ground state, one needs to account for states near to the ground state that give some contribution. There are different techniques (in the wein of Taylor expansions) to handle this in different situation, but AFAIK none of them work in every situation, and certainly aren't universal in the sense suggested in the article. So if there is a proof that all Ising models in the low energy regime are basically the same and can be solved with general these SAT-like techniques this is a new result.