DeepSeek's V4.1 Flash AI Model Outperforms Kimi K3 on Cybersecurity and Coding Benchmarks

DeepSeek, the Chinese artificial intelligence developer, announced on Thursday that its new V4.1 Flash model surpasses competitors on key benchmarks for cybersecurity and coding tasks. The model employs a novel "Causal-Encoder-Decoder" architecture and, despite being built on a 552 billion-parameter framework, uses a Mixture-of-Experts (MoE) design that activates only 8 billion parameters for input processing and 16 billion for response generation, significantly reducing computational demands per request. DeepSeek claims V4.1 Flash is the smallest in its new series and features native multimodal visual understanding. It outperformed the company's own V4 Pro on benchmarks evaluating coding, cybersecurity, and autonomous agent tasks. On Terminal-Bench 2.1, which assesses AI on real-world compute tasks, V4.1 Flash scored 90.6, beating OpenAI's GPT-5.6 Sol (88.8), Moonshot AI's Kimi K3 (88.3), and DeepSeek's V4 Pro (87.9). The launch underscores China's intensifying AI competition, where efficiency and speed are critical amid rising hardware costs and foreign chip export curbs that tighten compute constraints. Chinese developers are racing to deliver high-end reasoning at fraction-of-a-cent operational costs.