Thanks from reply Kaktor!
I would rather measure in MBs, that extra care, for a better result.
I hope that there will be some progress (perhaps there already is) in the field of compression in the near future.
From the above 1-2 results, it is clear that what was only available and approachable in 2011-2012 with slow CM compressors, LOLZ has brought to 2018, within the LZ family, together with fast unpacking. (WinRK: ~90-100 minutes vs. LOLZ less than 2-3 seconds.)
GDCC'23 also brought one or two interesting results, which may or may not have been useful for game textures and compression. I had a dataset of JPG images that contained a larger raw image and smaller size/quality "copies" of it, similar to textures found in games.
In terms of size, it was 940+ MB, even with precomp and other tricks it could not be reduced below 700MB, while the winning compressor was able to compress it by more than half, not much above the 300MB size. Losslessly. (PAQ8px_v208fix1 from T1 Dataset in has 670 million bytes.)
https://encode.su/threads/4048-GDCC2...recompression-
UPD: Previous post updated and probably finished more tests...