I wrote a "lite" version of paq8l called lpaq1, 35 times faster but files are 10% bigger (but still #8 on enwik9). It is a file compressor, not an archiver. http://cs.fit.edu/~mmahoney/compression/lpaq1.zip
I wanted something reasonably fast and simple and open source (GPL). It does not have any filters or specialized models (except 1 for text). It mixes only 7 contexts for speed (orders 1, 2, 3, 4, 6, a unigram word context, and a match model). It uses a neural mixer and 2 SSE stages. Source documentation explains how it works.