nvidiaCertifiedReleased 06/04/2026

Nemotron 3 Ultra

A 550B parameter (55B active) open reasoning model from NVIDIA, built for long-running agent workflows. It uses a hybrid Mamba-Transformer MoE architecture and supports a 1M token context window.

View API reference ↓
Input / output priceFrom$0.50 / $2.20per 1M
24h uptime99.98%
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.inferencebureau.com/v1",
  apiKey: process.env.IB_API_KEY,
});

const res = await client.chat.completions.create({
  model: "nvidia/nemotron-3-ultra-550b-a55b",
  messages: [{ role: "user", content: "Why is the sky blue?" }],
});

console.log(res.choices[0].message.content);

Providers

Dispatch routes requests across these providers and fails over between them without changing the model. Restrict or order providers with team Directives. Prices are gateway list prices before the platform fee.

ProviderContextMax outputLatencyThroughputInputOutputCacheCapabilitiesZDRNo trainingRegionsStatus
Baseten1M65K0.2 s261 tps$0.60$2.40Read $0.12Write —4 featuresYesYesUSCertified
DeepInfra262K65K0.4 s175 tps$0.50$2.20Read $0.100Write —5 featuresNoYesGlobalCertified

Uptime

24 hours

Request success rate per provider.

Baseten99.96%
DeepInfra100.00%

Throughput

last hour

P50 output tokens per second.

Baseten261 tps
DeepInfra175 tps

Latency

last hour

P50 time to first token.

Baseten0.2 s
DeepInfra0.4 s

Getting started

Create a key on the API Keys page and set it as IB_API_KEY. See the quickstart.

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.inferencebureau.com/v1",
  apiKey: process.env.IB_API_KEY,
});

const res = await client.chat.completions.create({
  model: "nvidia/nemotron-3-ultra-550b-a55b",
  messages: [{ role: "user", content: "Why is the sky blue?" }],
});

console.log(res.choices[0].message.content);

Top-level parameters

Standard Chat Completions parameters work as documented. These have model-specific behavior.

ParameterTypeRequiredDescription
modelstringYesModel ID in the form creator/model, e.g. nvidia/nemotron-3-ultra-550b-a55b. Must be allowed by your team's Directives.
max_tokensnumberNoHard cap on generated tokens, and the basis for the balance reserved per request. Nemotron 3 Ultra supports up to 65,000 output tokens.
reasoning{ effort: 'low' | 'medium' | 'high' }NoReasoning effort, mapped to the provider's native configuration. Reasoning tokens count toward output usage.
toolsTool[]NoFunction definitions the model can call. Tool calls are returned in choices[].message.tool_calls.
Supportedmax_tokenstemperaturestoptoolstool_choicereasoninginclude_reasoningresponse_formatstructured_outputs

Input limits

InputSourcesLimits
Text—Prompt and response share the 1M-token context window

More models by nvidia

ModelContextInputOutputCacheRelease
nvidia/nemotron-3.5-super-vl1.05M$0.20$0.50Read $0.040Write —10/15/2026
nvidia/nemotron-3.5-lightning-free1M$0$0Read —Write —10/14/2026
nvidia/nemotron-3.5-lightning262K$0.060$0.16Read $0.030Write —08/11/2026
nvidia/nemotron-3-super-120b-a12b256K$0.15$0.65Read —Write —03/11/2026
nvidia/nemotron-nano-12b-v2-vl131K$0.20$0.60Read —Write —10/28/2025
nvidia/nemotron-nano-9b-v2131K$0.060$0.23Read —Write —08/18/2025