Nemotron 3 Ultra 550B A55B

Largest Nemotron 3 model for maximum open-weight reasoning and agent accuracy

Strongest at
Writing
Top 23% of the past year's models · one benchmark so far
Price per 1M tokens
$0 / $0
input / output · NVIDIA
Context window
1M
tokens in · 128K out

Capability profile

Scored against 91 models released in the past year. 50 is the average model; further out is better.

This modelMedian of the past year
5090Median of the past yearCoding: 27 · Better than 6%Writing: 62 · Top 23%Reading & documents: 13 · Better than 1%Reasoning: 30 · Better than 5%Math: 50 · Better than 29%Science: 48 · Better than 6%CodingWritingReadingReasoningMathScience
Coding27Better than 6% · 1 test
Writing62Top 23% · 1 test
Reading & documents13Better than 1% · 1 test
Reasoning30Better than 5% · 1 test
Math50Better than 29% · 1 test
Science48Better than 6% · 1 test

No results yet: Visual reasoning, Computer use

Where these scores come from

Each capability combines the published benchmarks listed here, rescaled so 50 is the reference average and 15 points is one standard deviation. A hollow dot rests on a single benchmark.

NVIDIA's past year

13 release days in the past year, about one every 28 days. The next is expected around Nov 19, 2026. Forecasts ↗

NovDecJanFebMarAprMayJunJulAugSepOctNov

Specifications

RELEASE DATEJun 4, 2026
WEIGHTSOpen weights
CONTEXT WINDOW1M
MAX OUTPUT128K
REASONINGNot listed
TOOL CALLINGNot listed
INPUTStext
OUTPUTStext

Models.dev record ↗

Specifications and prices from Models.dev; availability and limits may differ by provider. Benchmarks from Epoch AI (CC BY 4.0), LiveBench and the other sources named above.