I've always been told that it's the other way around. Every vertex and triangle in a singular model is rendered in batch, whereas each brush face in a map is rendered with a separate draw call. This means that when comparing a bunch of brushes to a single model, both forming the same shape, the model will be cheaper to render.
In practice, this difference isn't something the average mapper needs to worry about, unless you're working with
lots--and I mean
lots--of vertices. On a small scale as with a single cube, the difference is practically nonexistent. There are many more effective ways to optimize a map, which are discussed pretty extensively in
this guide if you're interested. The primary way to optimize a map isn't to make it faster to render what's visible, but to reduce what's visible in the first place, as the most expensive objects in the average TF2 match are other players and the things they bring with them (projectiles, cosmetics, particles, etc). Additionally, as iiboharz pointed out, many models reach a level of detail that is difficult to recreate with brushes, so something like a simple brush window might be more expensive than a well-detailed prop window. But again, this won't really create a noticeable difference unless you're absolutely spamming props.