N
Nemotron 3 Ultra
Scores reviewed 2026-06-27Prices checked 2026-10-25NVIDIA
- OpenRouter
- nvidia/nemotron-3-ultra-550b-a55b
- Context
- 262K
- Price per 1M
- $0.6 in / $2.4 out
- Tooling
- 4/5
- Deploy
- Cloud API
Monthly cost, worked example
| Input | 667,000 tokens × $0.6 per 1M | $0.4 |
|---|---|---|
| Output | 333,500 tokens × $2.4 per 1M | $0.8 |
| Total | ~$1.2/mo |
667 requests a month at 1,000 input and 500 output tokens each, about 1M tokens. The recommender uses the same assumption when you don't give your own usage.
Best for
- Agentic workflows and tool calling
- Writing and editing code
- Complex reasoning and analysis
- Long documents (~262K context)
Skip if
- You need open weights to self-host
Capability scores
Tool calling4/5
Tooling (1-5): how reliably the model calls tools and follows multi-step instructions. Editorial score from LMFinder, not a benchmark.
Multi-step reasoning4/5
Specifications
- Context window
- 262K tokens
- VRAM
- N/A (cloud-hosted)
- License
- proprietary
- Pricing
- $0.6 in / $2.4 out per 1M
- Deployment
- Cloud API
- Docs
- Model docs ↗
Our take
Nemotron 3 Ultra from NVIDIA offers a 262K-token context window and is well-suited to strong tool calling, multi-step reasoning, coding, long-context work. Pricing is $0.6 in / $2.4 out per 1M, billed per token through the provider API.
Watch for the trade-offs: weights are closed, so you cannot self-host.
Compare to
Not sure this is the one?
Describe your project and let LMFinder rank the field for your exact constraints.
Describe your project →