NVIDIA’s DeepSeek-V4.1-Flash-1017-NVFP4 is a quantized, multimodal MoE language model for agent systems, chatbots, RAG, and other AI applications, based on DeepSeek-V4.1-Flash-1017 and using a Causal Encoder-Decoder architecture with Compressed Sparse Attention 2. It is maintained by nvidia under an MIT license, with commercial and non-commercial use described. Use cases include text and image input with text generation, chatbot and RAG applications, and AI-agent deployments including tool calling through the documented vLLM configuration. The base model may produce biased or toxic responses, inaccuracies, omissions, or irrelevant and redundant text; the checkpoint requires a vLLM build with MXFP4 Engram support, stock vLLM images cannot load its Engram tables, and SGLang has not been validated. The related DeepSeek-V4.1-Flash-NVFP4 is a different version whose checkpoint or results should not be assumed interchangeable. Against the stated MXFP4 source baseline, reported scores were 92.93 vs. 92.77 on GPQA Diamond, 80.19 vs. 80.41 on AA-LCR, 57.32 vs. 57.58 on SciCode, 76.47 vs. 76.33 on IFBench, 78.73 vs. 78.38 on MMM