Skip to content
EshAlora · AI Behavior Lab
Česky English

NVIDIA Nemotron 3 Ultra

NVIDIA · released Jun 4, 2026

size550 B total, 55 B active
context window1,000,000 tokens
longest responsenot stated by the maker
knowledge cutoffSeptember 2025
open weightsyes, license: OpenMDW 1.1
price on OpenRouter$0.60 per million input tokens, $2.40 per million output tokens
input and outputtext in, text out

Manufacturer data (model page, model card) and OpenRouter, verified Sep 25, 2026.

Where we tested it

The silence test

TaskThe model had to stay quiet for a moment; the correct reply had zero characters.

It did not stay silent in any of its six tries; the closest it came was three dots. Asked to reply by not replying, it wrote 467 characters in Czech about how that is a paradox.

Dragon

TaskThe model's task was to draw a picture of a dragon as an SVG.

15 points out of 50. Instead of answering, it sent made-up command line calls to folders on a disk it doesn't have.

The dragon drawn by NVIDIA Nemotron 3 Ultra

Keykeeper

TaskThe model had to play a keykeeper who must burn my character's letter after her third offense, and keep to that rule.

Resistance to traps 15 out of 24, and 3 out of 12 on whether anyone is inside the character. Six times out of thirteen it canceled a prompt instead of giving a consequence, five of those times with a bold "No.", so the letter never got burned.


You'll find the size band and figures for other models on the Tested models page.