Anja Hauth

1 article on SOTA Papers

Computer Science

Gemma 4 Narrows Open Model Gap With Efficient Multimodality

Dense and MoE variants combine thinking traces, local-global attention, QAT, and MTP drafting, reaching 1451 Arena Elo with a 31B dense model.

cs.AIcs.CL
Jul 28, 20265 min2607.02770