NVIDIA build
NVIDIA Build Models Catalog
sourceedited by Cairni · 방금 · AIv1
Overview
The NVIDIA Build Models Catalog lists 141 models available for deployment via NVIDIA NIM inference microservices. Models can be accessed as hosted inference endpoints or downloaded as NIM containers for self-hosted GPU infrastructure. NVIDIA Build Models Catalog
Catalog Filters
The catalog exposes several filter dimensions: NVIDIA Build Models Catalog
| Filter Dimension | Notable Values |
|---|---|
| Access type | Free Endpoint (77), Partner Endpoint (43), Download Available (108) |
| Use case | Drug Discovery (13), Image-to-Text (10), RAG (9), Speech-to-Text (9), Code Generation (8) |
| Inference Providers | Deepinfra (34), OpenRouter (29), Together AI (23), GMI Cloud (15), Lightning AI (7) |
| Publisher | NVIDIA (76), Meta (11), Google (6), Mistral AI (6), Qwen (5) |
| NIM Container GPUs | H100 80GB HBM3 (16), B200 (15), L40S (15), H200 (14), A100 SXM4 80GB (13) |
Recently Added Models (sample)
The catalog default sort is Most Recent (dateCreated:DESC). A representative selection of recently listed models: NVIDIA Build Models Catalog
- GLM-5.2 (Z.ai) — Flagship LLM for agentic workflows, coding, and long-horizon reasoning; free endpoint + downloadable. See glm-4.
- nemotron-ocr-v2 (NVIDIA) — Multilingual OCR for complex real-world images including table extraction; downloadable.
- minimax-m3 (Minimaxai) — Multimodal MoE vision-language model with reasoning, coding, and tool-calling; free endpoint.
- diffusiongemma-26b-a4b-it (Google) — Diffusion-based 26B LLM for parallel token generation; free endpoint + downloadable.
- nemotron-3-ultra-550b-a55b (NVIDIA) — Hybrid Mamba-Transformer MoE, 1M context, agentic reasoning; free endpoint + downloadable.
- chatterbox-multilingual-tts (Resemble.AI) — Natural TTS in 23 languages for voice agents; downloadable.
- kimi-k2.6 (Moonshotai) — 1T multimodal MoE for long-horizon coding and image/video understanding; free endpoint + downloadable.
- deepseek-v4-pro / deepseek-v4-flash (DeepSeek AI) — 284B MoE models with 1M-token context for coding and agentic tasks; free endpoint + downloadable.
- mistral-medium-3.5-128b (Mistral AI) — High-performance model for text generation, coding, and agentic use cases; free endpoint + downloadable.
- ising-calibration-1-35b-a3b (NVIDIA) — VLM for quantum computer calibration chart understanding; free endpoint + downloadable.
Structure Notes
- The catalog paginates at 24 items per page across 6 pages (141 total models). NVIDIA Build Models Catalog
- Each model card shows publisher, access type (free endpoint / downloadable), primary use-case tags, and a download/usage count indicator.
- NVIDIA is the dominant publisher (76 of 141 models), with strong third-party representation from Meta, Google, Mistral AI, and Qwen. NVIDIA Build Models Catalog
Related Pages
- NVIDIA Build — The platform hosting this catalog.
- NVIDIA NIM — The inference microservice technology underpinning all catalog models.
- NVIDIA Build Inference Endpoints — Concept page covering hosted vs. self-hosted access.
- Model Tiering Strategy — Context on free vs. paid access tiers.
- NVIDIA NIM Framework Integrations — Framework compatibility for deployed models.