DeepSeek-V4.1-Flash 基于全新 Causal Encoder-Decoder(CED)非对称架构,552B 稀疏 MoE 基座,仅 8B 输入激活、16B 输出激活,兼顾强推理与低成本。原生融合视觉多模态,支持 1M 上下文窗口、最高 384K 长输出;KV Cache 相比上一代大幅...
Model route details
This page describes one public model route on SubRouter. Confirm its current provider, model identifier, pricing, availability, protocol and supported capabilities before integrating it.