Curated Model Index

Only proven smaller-than-MLX wins.

This page no longer lists every recent upload. It only surfaces releases tied to a decisive benchmark or coherency win where JANG is smaller than the compared MLX baseline.

MiniMax-M2.5-JANG_2L

82.5 GB, 2.10 bits, 74% MMLU. Beats MLX 4-bit by +47.5 points while being 37.3 GB smaller.

Open full JANGQ-AI account →