You can do the following:

• compare the overall fit of the models (chi-square test)
• look at information criteria (e.g., AIC and BIC values)
• some models may be nested so you may be able to conduct chi-square difference tests under certain circumstances

That being said, I generally discourage the use of second-order CFA models as in my opinion they rarely make sense for empirical applications from a measurement theoretical point of view. An exception are measurement designs in which the first-order factors are interchangeable (randomly selected from a universe of exchangeable constructs/facets). This may be the case for designs with multiple randomly selected raters and for longitudinal designs with randomly selected measurement occasions/situations. In most other cases, the first-order factors tend not to be interchangeable and therefore not in line with the sampling design implied by hierarchical factor models. See

Eid, M., Geiser, C., Koch, T., & Heene, M. (2017). Anomalous results in G -factor models: Explanations and alternatives. Psychological Methods, 22(3), 541–562. https://doi.org/10.1037/met0000083

Christian Geiser thank you very much for your help.