Nemotron 3 Ultra (free) API
NVIDIA's Nemotron 3 Ultra is an open 550B-parameter Mixture-of-Experts reasoning model with a 1-million-token context, built for agentic workflows and deep research.
Nemotron 3 Ultra is NVIDIA's final and best model in the Nemotron 3 family: an open frontier-reasoning and orchestration model built on a Hybrid Transformer-Mamba mixture-of-experts architecture. It has 550 billion total parameters with 55 billion active parameters. It uses LatentMoE for improved accuracy and includes MTP layers for faster inference through native speculative decoding. The model was pretrained in NVFP4 format and post-trained with Supervised Fine Tuning, Reinforcement Learning, and Multi-teacher On-Policy Distillation. It takes text input and returns text output, with a 1 million token context window.
The model is designed for multi-step reasoning and planning, agent orchestration, coding agents, deep research, and complex enterprise tasks. It supports function calling through tools and tool_choice parameters, structured outputs via a JSON schema in response_format, and inference time reasoning budget control. NVIDIA positions it for long-running autonomous agent workflows and high-throughput inference in high-volume agent pipelines. It outperforms state-of-the-art open LLMs on RULER at 1M context length. The free endpoint is rate limited.
NVIDIA releases open weights, training data, and recipes, including pre-trained, post-trained, and quantized checkpoints along with training datasets. On OpenRouter the model is served by three providers: DeepInfra, Baseten, and Venice.
Updated:
MODEL ID
nemotron-3-ultra-550b-a55b:free- Context
- 1M
- Max output
- 65.536K
- Recommended API
- messages
- Input
- text
- Output
- text
Highlights
- 550B total, 55B active parameters
- Hybrid Transformer-Mamba MoE architecture
- 1 million token context window
- Open weights, training data, and recipes
- Built for high-throughput agent pipelines
- Free endpoint is rate limited
Providers
Compare API prices across providers. Select a provider to view all charges and conditions.
| Uptime | |||
|---|---|---|---|
| $0 | $0 |
Historical observations are not a live health check. Missing data does not mean an outage.
Benchmarks
- 48
Artificial Analysis
Score reported in a single review
Scores as published by the model provider.
Reasoning
- Default effort
- high
- Enabled by default
- Yes
- Always enabled
- No
- Available effort levels
- high / medium / none
Quick Start
Drop-in code to call this model with Felo API Platform compatible APIs. Start in the Playground when you want to test prompts before writing code.
Get your API key
Create a key from your Felo API Platform account and set it as an environment variable.
export FELO_API_KEY="YOUR_API_KEY"Make your first request
Use this model ID with the Felo API Platform endpoint. Optional request headers such as X-Request-Id can help troubleshooting.
nemotron-3-ultra-550b-a55b:freeThis model is reasoning-capable. If your selected protocol and model route expose reasoning controls, preserve the returned reasoning metadata when continuing a conversation.
const response = await fetch("https://openapi.felo.ai/api/v1/messages", {
method: 'POST',
headers: {
Authorization: `Bearer ${process.env.FELO_API_KEY}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
"model": "nemotron-3-ultra-550b-a55b:free",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
]
}),
});
if (!response.ok) throw new Error(await response.text());
console.log(await response.json());Using third-party SDKs
Use compatible SDKs by setting baseURL to https://openapi.felo.ai/api/v1. Use the Anthropic-compatible base URL for Claude-style clients. https://openapi.felo.ai/api.
Enable streaming
Add stream: true to receive responses as server-sent events.
curl -N https://openapi.felo.ai/api/v1/messages \
-H "Authorization: Bearer $FELO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "nemotron-3-ultra-550b-a55b:free",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
],
"stream": true
}'Endpoint
Felo API Platform exposes compatible model-calling surfaces. Pick the surface your app or agent client already supports.
Messages
Anthropic-compatible messages for Claude-style clients.
POST
https://openapi.felo.ai/api/v1/messagesParameters
modelstringnemotron-3-ultra-550b-a55b:freePublic model ID to call.max_tokensinteger—Maximum generated tokens.messagesarray—Conversation messages for Chat Completions.seed———toolsarray—Tool calling definitions following compatible schema.top_p———reasoning———temperaturenumber—Controls response variety when supported.tool_choicestring or object—Controls whether a tool should be called.reasoning_effort———include_reasoning———Frequently asked questions
What is Nemotron 3 Ultra (free)?
Nemotron 3 Ultra is NVIDIA's final and best model in the Nemotron 3 family: an open frontier-reasoning and orchestration model built on a Hybrid Transformer-Mamba mixture-of-experts architecture. It has 550 billion total parameters with 55 billion active parameters. It uses LatentMoE for improved accuracy and includes MTP layers for faster inference through native speculative decoding. The model was pretrained in NVFP4 format and post-trained with Supervised Fine Tuning, Reinforcement Learning, and Multi-teacher On-Policy Distillation. It takes text input and returns text output, with a 1 million token context window. The model is designed for multi-step reasoning and planning, agent orchestration, coding agents, deep research, and complex enterprise tasks. It supports function calling through tools and tool_choice parameters, structured outputs via a JSON schema in response_format, and inference time reasoning budget control. NVIDIA positions it for long-running autonomous agent workflows and high-throughput inference in high-volume agent pipelines. It outperforms state-of-the-art open LLMs on RULER at 1M context length. The free endpoint is rate limited. NVIDIA releases open weights, training data, and recipes, including pre-trained, post-trained, and quantized checkpoints along with training datasets. On OpenRouter the model is served by three providers: DeepInfra, Baseten, and Venice.
How is usage billed?
Your bill depends on the provider you use, its prices and your usage. Longer inputs or other conditions may have different prices.
What input and output modalities does Nemotron 3 Ultra support?
It accepts text input and produces text output. It is a text-only model with no other modalities listed.
How long is the context window?
Nemotron 3 Ultra supports a context of 1 million tokens. The factsheet does not specify a separate output limit.
What tasks is it suited for?
It is built for coding agents, deep research, complex enterprise workflows, and long-running autonomous agents. It also handles multi-step reasoning, planning, and agent orchestration.
Does it support tool use and structured outputs?
Yes. It supports function calling via tools and tool_choice parameters, and structured outputs via a JSON schema in response_format. It also offers inference time reasoning budget control.
How does it differ from other Nemotron 3 models?
It is described as the final and best model of the Nemotron 3 family. It combines a Hybrid Transformer-Mamba mixture-of-experts design with LatentMoE and MTP layers.
What are its limitations?
Free endpoints are rate limited. No other limitations are listed in the factsheet.
Related models
Nemotron 3.5 Super VL (free)
NVIDIAnemotron-3.5-super-vl:freeNVIDIA's multimodal Nemotron 3.5 Super VL handles text, images, documents, GUI screenshots, and video, combining visual understanding, reasoning, instruction following, and tool use in one model.
Context
1M
Max output
131.072K
Input tokens
$0 / 1M tokens
Output tokens
$0 / 1M tokens
Nemotron 3.5 Lightning (free)
NVIDIAnemotron-3.5-lightning:freeNVIDIA's open-weight, text-only Mixture-of-Experts model built as a fast execution engine for high-volume agentic workloads, with up to 1M token context.
Context
1M
Max output
65.536K
Input tokens
$0 / 1M tokens
Output tokens
$0 / 1M tokens