SU(N) Yang-Mills theory in three dimensions, with a Chern-Simons term of level k (an integer) added, has two dimensionful coupling constants, g 2 k and g 2 N; its possible phases depend on the size of k relative to N. For k ≫ N, this theory approaches topological Chern-Simons theory with no Yang-Mills term, and expectation values of multiple Wilson loops yield Jones polynomials, as Witten has shown; it can be treated semiclassically. For k = 0, the theory is badly infrared singular in perturbation theory, a non-perturbative mass and subsequent quantum solitons are generated, and Wilson loops show an area law. We argue that there is a phase transition between these two behaviors at a critical value of k, called k c , with k c /N ≈ 2 ± .7. Three lines of evidence are given: First, a gauge-invariant one-loop calculation shows that the perturbative theory has tachyonic problems if k ≤ 29N/12. The theory becomes sensible only if there is an additional dynamic source of gauge-boson mass, just as in the k = 0 case. Second, we study in a rough approximation the free energy and show that for k ≤ k c there is a non-trivial vacuum condensate driven by soliton entropy and driving a gauge-boson dynamical mass M, while both the condensate and M vanish for k ≥ k c . Third, we study possible quantum solitons stemming from an effective action having both a Chern-Simons mass m and a (gauge-invariant) dynamical mass M. We show that if M > ∼ 0.5m, there are finite-action quantum sphalerons, while none survive in the classical limit M = 0, as shown earlier by D'Hoker and Vinet. There are also quantum topological vortices smoothly vanishing as M → 0.