Ling-2.6-1T
Ant Group's open-weight, non-thinking flagship: ~1T parameters with 63B active, a hybrid MLA and Linear Attention stack, a 1M-token context, and an MIT license.