Take open AI from model to production on premises.
Discover validated models, deployable applications, and hardware-aware configurations engineered for secure AI on Dell infrastructure.
dell-ai brings catalog discovery, configuration, and deployment
to your terminal. Explore the CLI & SDK Latest optimized models
Recently added open models, ready for Dell infrastructure.
DeepSeek V4.1 Flash 1017
New Updated Mixture-of-Experts model with improved reasoning, coding, agentic capabilities, and visual understanding, with shorter outputs and faster task execution.
Google EmbeddingGemma 2
New Multimodal embedding model with 740M parameters that maps text, images, video, and audio into a shared 768-dimensional vector space for semantic search, retrieval, classification, and clustering.
Beam 1.0 501B A23B BF16
New Reflection’s BF16 MoE model with 501B total and 23B active parameters, a 1M-token context window, and reasoning, coding, and tool-use capabilities.
Beam 1.0 501B A23B NVFP4
New Reflection’s NVFP4 MoE model with 501B total and 23B active parameters, a 1M-token context window, and reasoning, coding, and tool-use capabilities.
Muse Spark Open 816B A42B NVFP4
New NVFP4 MoE model with 816B total and 42B active parameters for code generation, debugging, codebase understanding, and agent workflows.
NVIDIA Nemotron 3.5 Super VL 120B A12B BF16
New BF16 vision-language model with 120B total and 12B active parameters for reasoning, coding, tool use, and document and video understanding. Supports up to 1M tokens of context and native MTP speculative decoding.
Latest updates
New models, platform support, and practical deployment guidance.
DeepSeek V4.1 Flash 1017 and EmbeddingGemma 2 land on Dell Enterprise Hub
DeepSeek V4.1 Flash 1017 improves reasoning, coding, agentic tasks, and visual understanding while reducing task execution time and output length. Deploy it with vLLM v0.32.0 on NVIDIA H100, H200, and B300 or AMD MI300X and MI355X systems. Google EmbeddingGemma 2 brings text, image, video, and audio embeddings into a shared vector space for semantic search and retrieval. Deploy it with hf serve v0.2.0 on 1× NVIDIA L40S or Dell Pro Max with GB10. Deploy DeepSeek V4.1 Flash 1017 → · Deploy EmbeddingGemma 2 →
Introducing Reflection Beam 1.0 on Dell Enterprise Hub
Beam 1.0 by Reflection is now available in BF16 and NVFP4 on Dell Enterprise Hub. This MoE model combines 501B total and 23B active parameters with a 1M token context window for reasoning, coding, and agentic tool use. Reflection reports comparable reasoning quality to leading open models using approximately 3× fewer reasoning tokens. Deploy NVFP4 on 4× NVIDIA H100 or 4× and 8× H200 and B300, with DFlash speculative decoding on B300. Deploy BF16 on 8× H200 or B300. Deploy Beam 1.0 BF16 → · Deploy Beam 1.0 NVFP4 →
Now available: Muse Spark Open NVFP4 and NVIDIA Nemotron 3.5 Super VL
Muse Spark Open improves code generation, complex debugging, codebase understanding, and end to end developer workflows while retaining strong general agent capabilities. Deploy the NVFP4 variant with SGLang v0.5.21. NVIDIA Nemotron 3.5 Super VL combines reasoning, coding, and tool use with document and video understanding. Its BF16 release offers 120B total and 12B active parameters, up to 1M tokens of context, and native MTP speculative decoding. Deploy with vLLM v0.31.0. Deploy Muse Spark Open NVFP4 → · Deploy Nemotron 3.5 Super VL →
Xiaomi MiMo V2.6 Pro and Flash land on Dell Enterprise Hub
Xiaomi’s MiMo V2.6 Pro RL and MiMo V2.6 Flash RL are now available on Dell Enterprise Hub, bringing frontier intelligence on prem. These native omnimodal Mixture of Experts reasoning models combine text, image, video, and audio understanding with a 1M token context window and a DFlash style speculative decoder. Deploy Flash on 4× or 8× NVIDIA H200 and B300 systems, and Pro on 8× H200 or 4× and 8× B300 systems. Deploy MiMo V2.6 Flash RL → · Deploy MiMo V2.6 Pro RL →
Deploy complete AI applications
Go beyond endpoints with production-ready AI experiences.
Super Analyzer
Super Analyzer is an application that identifies and evaluates common anti-patterns in C++, Java, Python, and Rust code. It uses a multi-agent architecture with three coordinated agent types—Primary, Fixer, and Chat—to deliver accurate analysis and iterative improvements. The Web UI supports natural multi-turn interactions and uses a PostgreSQL database to persist in user accounts and conversation history across sessions. Super Analyzer also provides a REST/API layer and a Python interface, all secured through user/password authentication backed by PostgreSQL. Access is protected with JWTs signed using RSA-256, ensuring a consistent and secure experience across all entry points.
OpenWebUI
OpenWebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform designed to operate entirely offline. It provides a modern interface for interacting with various LLM runners including Ollama and OpenAI-compatible APIs. With built-in RAG (Retrieval Augmented Generation) capabilities, OpenWebUI offers a comprehensive solution for AI deployment and interaction in Kubernetes environments.
AnythingLLM
AnythingLLM is an open-source, all-in-one AI chat application with built-in RAG (Retrieval Augmented Generation) capabilities. It provides a user-friendly interface for document management, LLM integration, and AI agent creation - all without complex setup. The application supports both local and cloud-based deployments, making it ideal for teams and organizations seeking a private, customizable AI solution.
Agentic Smart Router
In an Agentic AI application, not all prompts are the same, different prompts need to have different requirements in terms of complexity of the prompt - some need reasoning, some need access to RAG system and some need tool calling capabilities. In this custom-built Agentic Smart Router, we demonstrate that capability by leveraging NVIDIA Agent Intelligence Toolkit, NVIDIA NIMs, NVIDIA LLM Router blueprint and stitching them together to build an Agentic Smart Router application.
Optimized for Dell platforms
From AI PCs to accelerated PowerEdge servers, deploy with configurations tested for the hardware you already run.
