However, in KSPI [1]: Consider the combination of a nuclear lightbulb with compressed air as the reaction mass: Surprisingly good midpoint qua TWR and ISP! If you install a compressor you can get some pretty interesting SSTOs out of it. Whether these would work IRL is a different question, of course. [2]
Nuclear lightbulb is of course highly theoretical, but scientifically/engineeringly plausible at the least. [3]