Silicon-based Flow's extensive model service platform has recently launched DeepSeek-V3.1, a model generously open-sourced by the DeepQuest team. Boasting a staggering total of 671 billion parameters, with 37 billion active parameters, DeepSeek-V3.1 employs a versatile hybrid inference architecture capable of supporting both cognitive and non-cognitive operations. Its groundbreaking innovation lies in being the pioneer to support an ultra-long context of 160K, rendering it exceptionally well-suited for intricate scenarios encompassing long document processing, multi-turn dialogues, coding, and agent-based applications.