A very un-scientific screen grab and saving as a GIF gives me 604kB at 990×1000 pixels, but this does include color fringing at the edges of each letter and I'd be inclined to treat this as a 1 BPP image.
I bet somebody in the biz has a pipeline for running different algorithms/formats/compressions, and applying different metrics for "would these look about the same to a human on a certain screen at a certain distance."
> 1 BPP
Yeah, the different ASCII characters might be best modeled/analogized as local dithering.