DeepSeek launches V4 series with 1.6T-parameter MoE model

Arxiv CS.CL (NLP/language models)

The preview includes two MoE models: DeepSeek-V4-Pro with 1.6T total parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated), both supporting million-token contexts. Community quantizations like GGUF are already available.

DeepSeekMixture of experts

See 1 more source

Bartowski has delivered DS4 GGUF4 hours agochallis88ocarinaDiscuss

Arxiv CS.CL (NLP/language models)·DeepSeek-AIAnyi XuBangcai LinBing XueBingxuan WangBingzheng XuBochao WuBowei ZhangChaofan LinChen DongChenchen LingChengda LuChenggang ZhaoChengqi DengChengyu HouChenhao XuChenze ShaoChong RuanConner SunDamai DaiDaya

Jun 19, 4:00 AM