---
title: "Nemotron 3 Ultra"
description: "NVIDIA's largest open Nemotron 3 model, 550B parameters, for agentic reasoning."
canonical: "https://vetted.tools/models/nemotron-3-ultra"
source: "https://vetted.tools/models/nemotron-3-ultra.md"
---

# Nemotron 3 Ultra

NVIDIA's largest open Nemotron 3 model, 550B parameters, for agentic reasoning.

Source: https://vetted.tools/models/nemotron-3-ultra

- Slug: `nemotron-3-ultra`
- By [NVIDIA](https://vetted.tools/companies/nvidia) ([nvidia.com](https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/), checked 2026-09-26)
- Official page: [https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/](https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/)

## Facts

- License: other ([Hugging Face](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16), checked 2026-09-25)
- Capabilities: Reasoning, Tool use, Structured output, Implicit caching ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Context window: 1M tokens (1,000,000) ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Gateway ID: nvidia/nemotron-3-ultra-550b-a55b ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Input: $0.60 per 1M tokens ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Max output: 65K tokens (65,000) ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Takes: Text ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Returns: Text ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- No training on prompts: Offered by every provider on the Vercel AI Gateway, when you turn it on ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Output: $2.40 per 1M tokens ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Released: Jun 4, 2026 ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Type: Language ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)
- Zero data retention: Available on a paid Vercel plan through the AI Gateway, from every provider on it (you turn it on) ([Vercel AI Gateway](https://ai-gateway.vercel.sh/v1/models), checked 2026-09-25)

## Usage

- 171,251 downloads on [Hugging Face](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16) over the 30 days to 2026-09-25.
- 11,202,617 requests and 219,590,878,388 tokens on [OpenRouter](https://openrouter.ai/nvidia/nemotron-3-ultra-550b-a55b) over the 31 days to 2026-09-25.

## Elsewhere on this site

- Home: https://vetted.tools/index.md
- Use cases: https://vetted.tools/use-cases.md
- Tools: https://vetted.tools/tools.md
- The brief: https://vetted.tools/brief.md
- Assessment: https://vetted.tools/assessment.md
- Site summary for agents: https://vetted.tools/llms.txt
- Every published page in one file: https://vetted.tools/llms-full.txt
- Every URL the site publishes: https://vetted.tools/sitemap.xml