[Guojin Securities: DeepSeek V4.1 Flash Released - Focus on Domestic Large Models and AI Applications] Guojin Securities' research report states that DeepSeek V4.1 Flash was released on September 10, featuring native multimodality, 1 million tokens of context, and 384K output. It employs asymmetric activation (input 8B/output 16B), reduces cache to 890 bytes/token, and lowers VRAM usage to one-quarter of the previous generation. API prices have dropped significantly, but the model weights are 510GB, requiring an 8-GPU node for local deployment, leading to an increase in TCO (Total Cost of Ownership). Among seven tests, it outperformed V4 Pro in five areas related to coding and intelligent agents, though it lagged in two inference-related aspects. In financial research and investment: valuation models can be recalculated, and industry research entity pages rank second only to GPT-6 Astra, with strict backtesting standards but lacking trading constraints. Focus on domestic large models and AI applications.