NIM Use-Case Taxonomy

The NVIDIA Build model catalog organises its 141+ NVIDIA NIM models under a set of use-case tags. These tags act as the primary taxonomy for discovery: every model carries at least one tag, and many carry several. Understanding the taxonomy helps engineers select the right model tier and deployment path — see also Model Tiering Strategy and NVIDIA Build Inference Endpoints. [NVIDIA Build Models Catalog]

The following categories appear explicitly in the catalog's filter panel. [NVIDIA Build Models Catalog]

The catalog UI shows five primary use-case filters with a "Show more" control exposing additional tags. The total catalog stands at 141 models at time of capture. [NVIDIA Build Models Catalog]

Many models carry multiple tags, reflecting cross-cutting capabilities. The table below lists notable multi-tag models visible in the catalog snapshot. [NVIDIA Build Models Catalog]

Beyond the five highlighted filters, the catalog snapshot reveals several niche or emerging tags applied to individual models. [NVIDIA Build Models Catalog]

• Synthetic Data Generation — e.g. qwen-image-edit-nvpcb-ovsl2sl, specialised for Omniverse synthetic-to-photographic conversion at NVIDIA PCB inspection stations.
• Diffusion LLM — e.g. diffusiongemma-26b-a4b-it, a diffusion-based 26B-parameter LLM for parallel token generation.
• LLM Safety — e.g. nemotron-3.5-content-safety and nemotron-3-content-safety, multilingual multimodal models for detecting unsafe content.
• Broadcast — e.g. Active Speaker Detection, LipSync, and synthetic-video-detector, covering AI-generated video detection and live-production tools.
• Quantum — e.g. ising-calibration-1-35b-a3b, an open VLM for quantum computer calibration chart understanding.
• Autonomous Vehicles / Physical AI — e.g. cosmos3-nano, generating physics-aware videos for physical AI development.
• Text-to-Image — e.g. qwen-image and qwen-image-edit, foundation and editing models with multilingual text rendering.
• TTS (Text-to-Speech) — e.g. chatterbox-multilingual-tts, supporting 23 languages for voice agents.
• HDRI / Relighting — e.g. Relighting, re-illuminating people in video using 360° HDRI environment maps.
• Table Extraction / OCR — e.g. nemotron-ocr-v2, robust multilingual OCR on complex real-world images.

*Source: [NVIDIA Build Models Catalog] — top five use-case filter counts as shown in the catalog UI.*

The diagram below shows how use-case tags relate to the two primary access modes on NVIDIA Build. [NVIDIA Build Models Catalog]

Several use-case tags map directly to compound AI system patterns:

• Agentic AI / Agent tags align with the Agentic Skills capability layer, covering models like GLM-5.2 and nemotron-3-ultra-550b-a55b that are optimised for long-horizon reasoning and tool calling.
• Retrieval Augmented Generation (9 models) underpins Blueprint Collection reference architectures.
• Code Generation models (8 models) are frequently paired with OpenAI-Compatible API endpoints for drop-in developer tooling.

For guidance on choosing between free-endpoint and downloadable tiers within these categories, see Free vs. Paid AI Model Access and the NVIDIA NIM Free Models Guide.