Both nice.
Well for starters Torso is designed and markets the S-4 as a “sculpting sampler”.
The GR-2 ( the GR-MEGA is in a whole other price category, so i’ve dropped it from this comparison ) has a sampler engine too, plus the granular engine, with a tape engine ( which is a variation on the sampler engine ), and a spectral engine.
I think a stark contrast is just in horseposer. The GR-2 has sixteen voices polyphony per layer, with total of four layers, and 128 grains per voice, with up to 4000 grains total.