Long-Context Operations

DeepSeek Open-Releases V3.2-Exp, a Sparse-Attention Model Claiming Up to 50% Lower Long-Context Costs

The open-weight release invites independent testing with day-one support for Chinese NPUs.