Nemotron 3 Ultra APITESTEDOPEN WEIGHTS1M context
NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE).
Pricing & specs mirror our live pricing as of July 2026 (pay-as-you-go). Coding score is our own measured result.
✅ First-party tested — we ran Nemotron 3 Ultra on our 9-task coding suite
tested on free tier. Executed against hidden tests at real billed cost — full results · how we test.
What is Nemotron 3 Ultra?
NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it...
Nemotron 3 Ultra on the release timeline
Nemotron 3 Ultra pricing
Pay-as-you-go on DataLLM Lab — these are our live list prices (identical to the pricing page):
| Model | Input / 1M | Output / 1M | Cache read | Cache write | Context |
|---|---|---|---|---|---|
| Nemotron 3 Ultra | $0.50 | $2.20 | $0.10 | — | 1M |
On output price, Nemotron 3 Ultra is cheaper than 40% of the 309 models in the catalog.
What Nemotron 3 Ultra costs per month
| Workload | Tokens in / out (monthly) | Est. cost |
|---|---|---|
| Support chatbot | 40M / 12M | $46.40 |
| RAG / knowledge base | 200M / 20M | $144 |
| Coding agent | 80M / 25M | $95.00 |
| Batch extraction | 150M / 8M | $92.60 |
| Content generation | 20M / 40M | $98.00 |
Estimate your own monthly cost
Cost = input price × input volume + output price × output volume. The same five workloads run on every model page, so any two compare directly.
Nemotron 3 Ultra vs alternatives
| Model | Input / 1M | Output / 1M | Context | Our test |
|---|---|---|---|---|
| Nemotron 3 Ultra | $0.50 | $2.20 | 1M | 8/9 |
| Llama 3.3 Nemotron Super 49B V1.5 | $0.40 | $0.40 | 131K | — |
| Nemotron 3 Super | $0.08 | $0.40 | 1M | — |
| Nemotron 3 Nano 30B A3B | $0.05 | $0.20 | 262K | — |
| MiniMax M1 | $0.40 | $2.20 | 1M | — |
| GLM 4.5 | $0.60 | $2.20 | 131K | — |
When not to use Nemotron 3 Ultra
- It is a heavy reasoner (~9,136 reasoning tokens across our 9 tasks) — slower and pricier per call; avoid it for high-volume, simple requests where a lean model is far cheaper.
Specs
| Model ID | nvidia/nemotron-3-ultra-550b-a55b |
| Modality | text->text (input: text) |
| Context window | 1,000,000 tokens |
| Max output | 16,384 tokens |
| Tool / function calling | ✅ Yes |
| Structured output (JSON) | ✅ Yes |
| Released | 2026-06-04 |
| Open weights | ✅ nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 |
How to call Nemotron 3 Ultra
from openai import OpenAI
client = OpenAI(base_url="https://api.datallmlab.com/v1", api_key="YOUR_DATALLM_LAB_KEY")
resp = client.chat.completions.create(
model="nvidia/nemotron-3-ultra-550b-a55b",
messages=[{"role": "user", "content": "Hello"}],
# supports tools=[...] and tool_choice
)
print(resp.choices[0].message.content)curl https://api.datallmlab.com/v1/chat/completions \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"nvidia/nemotron-3-ultra-550b-a55b","messages":[{"role":"user","content":"Hello"}]}'import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://api.datallmlab.com/v1", apiKey: process.env.DATALLM_KEY });
const r = await client.chat.completions.create({
model: "nvidia/nemotron-3-ultra-550b-a55b",
messages: [{ role: "user", content: "Hello" }],
});
console.log(r.choices[0].message.content);The same key routes Nemotron 3 Ultra and 300+ other models — switch models by changing one string. How OpenAI-compatible APIs work.
Prompting tips for Nemotron 3 Ultra
- State the goal, not every step. Nemotron 3 Ultra reasons internally — give it the objective, constraints and success criteria and let it plan the approach; over-scripting each step tends to lower quality. Turn effort up for hard problems.
- Huge 1M context — but anchor the ask. You can paste whole documents or codebases; models attend most to the start and end, so put the key instruction at the top and restate it after long inputs.
- Give it real tools. Pass a
tools=[...]schema instead of asking it to "pretend" — let Nemotron 3 Ultra emit tool calls, execute them, and feed results back for the next turn. - Constrain JSON with a schema. Use
response_format/ structured outputs rather than "reply in JSON" to get valid, parseable objects every time.
Tips are derived from Nemotron 3 Ultra's actual capabilities — context window, tool & JSON support, modality and price tier — not generic advice.
Open-source tools for Nemotron 3 Ultra
Popular open-source projects for running and building with Nemotron 3 Ultra — star counts pulled from GitHub (July 2026).
Listed by GitHub stars; inclusion is by ecosystem relevance (inference engines, agent frameworks and SDKs), not affiliation. Stars change — see each repo for current numbers.
Migrating to Nemotron 3 Ultra
Coming from OpenAI or another gateway? On an OpenAI-compatible setup the only changes are base_url → https://api.datallmlab.com/v1 and model → nvidia/nemotron-3-ultra-550b-a55b. Messages, streaming, tool calls and the rest of your code stay the same. Routing & failover guide.
Self-host or use the API?
Nemotron 3 Ultra ships open weights (nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16), so you can run it yourself. Most teams still use the API: no GPU to provision or keep warm, no inference ops, and instant access across multiple providers with automatic failover. Self-host when data residency or fixed per-token economics matter most.
Rate limits & reliability
Rate limits here are the DataLLM Lab gateway's, not the upstream vendor's. On 429 (rate limited) or 503 (provider busy), retry with exponential backoff. See the error-code guide and failover setup.
Related reading
Call Nemotron 3 Ultra with one key
300+ models behind one OpenAI-compatible endpoint — better prices, better uptime, no subscriptions.
Get an API keyCompare pricingFrequently asked questions
What is Nemotron 3 Ultra?
NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). It accepts text input with a 1M-token context window and was released on 2026-06-04. On DataLLM Lab it is callable through one OpenAI-compatible endpoint.
How much does Nemotron 3 Ultra cost?
On DataLLM Lab it is $0.50 per 1M input tokens and $2.20 per 1M output tokens, with cached input at $0.10/1M — cheaper than about 40% of the 309-model catalog on output price. Pay-as-you-go, no subscription.
What is the context window of Nemotron 3 Ultra?
1M tokens, with up to 16K max output tokens.
Is Nemotron 3 Ultra open source?
Yes — open weights are published on Hugging Face (nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16). You can self-host it or call it via DataLLM Lab.
How do I call the Nemotron 3 Ultra API?
Point any OpenAI SDK at https://api.datallmlab.com/v1 and set model to "nvidia/nemotron-3-ultra-550b-a55b". One DataLLM Lab key routes this model and 300+ others; no code changes beyond the base URL and model string.
How did Nemotron 3 Ultra score in real testing?
In our executed 9-task coding benchmark it scored 8/9, averaging 48.2s per task at ~$0.00 per 1,000 tasks (real billed cost), generating 9,136 reasoning tokens. See our methodology for scope and limits.