← Back to all models

DeepSeek V3.2 671B

DeepSeek family · DeepSeek

671B
deepseek-v3-2-671b architecture diagram
Architecture diagram
Parameters
671B
Context Length
200K
Released
Jan 2025
Size Class
Frontier
MoE MLA

DeepSeek's latest MoE model with Multi-Latent Attention for superior efficiency.

Sources

GitHub →