X @外汇交易员

Model Release and Architecture - DeepSeek officially released and open-sourced the DeepSeek-V3.2-Exp model on September 29 [1] - The model introduces a sparse Attention architecture, effectively reducing computing resource consumption and improving model inference efficiency [1] Deployment and Infrastructure - The model is now available on Huawei Cloud's Model as a Service (MaaS) platform [1] - Huawei Cloud continues to use the large EP parallel scheme for DeepSeek-V3.2-Exp deployment [1] - The deployment leverages a context-parallel strategy with long sequence affinity based on the sparse Attention structure, balancing model latency and throughput performance [1] Pricing and Cost Reduction - DeepSeek has significantly reduced the cost of the new model service, leading to a corresponding decrease in official API prices [1] - Developers can expect a cost reduction of over 50% when calling the DeepSeek API under the new pricing policy [1]