ScaledThought
← All models

NVIDIA

Nemotron 3 Ultra

Live · US + CAtextOpenMDW-1.1

NVIDIA's largest Nemotron 3: a hybrid Mamba-Transformer MoE tuned for agents and tool calling.

Price per 1M tokens

Input

$0.60

Cached input

$0.12

Output

$2.40

Priced per model · Same in US and CA · Free tier included · Pricing

AA Index

23

#12 of 14
open models we serve

Specs

Model id
nemotron-3-ultra-550b-a55b
Weights released
Jun 4, 2026
Parameters
550B total · 55B active
Architecture
MoE
Context
—
Input
text
License
OpenMDW-1.1
Regions
us, ca

Benchmarks

  • Artificial Analysis Intelligence Index23
  • GPQA Diamond86.7%
  • AIME 202582.71%

Sources: huggingface.co · research.nvidia.com · artificialanalysis.ai

Use it

Three lines of code, or zero.

OpenAI-compatible. Use the model id below, or auto to let the router decide when this model is worth it.

Quickstart→
import os
from openai import OpenAI

client = OpenAI(base_url=os.environ["ST_BASE_URL"], api_key=os.environ["ST_API_KEY"])
resp = client.chat.completions.create(
    model="nemotron-3-ultra-550b-a55b",
    messages=[{"role": "user", "content": "Hello"}],
)

Opens your agent with the prompt typed. You press Enter.