> Nemotron 3 Ultra is an open-weights AI model from NVIDIA, released 4 Jun 2026. Index score 145 (GPT-4 = 100), rank 73 of 229 ranked models. API price $0.60 in / $2.4 out per million tokens, 262K context.

[Models](https://themodelindex.org/models/) /[NVIDIA](https://themodelindex.org/labs/nvidia/)

# Nemotron 3 Ultra

NVIDIA

Released 4 Jun 2026

Open

Date verified

[Compare](https://themodelindex.org/compare/nemotron-3-ultra-vs-claude-fable-5/) [Run it yourself →](https://themodelindex.org/local/) [Announcement](https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/)

145133–157

Index score

#73 of 229 models

1/ 8

Core tests taken

1 of 6 areas · plus 8 anchor tests

−32

Versus the frontier at release

best then: GPT-5.5

$0.60/ $2.4

Price per million tokens

input / output, list price

262K

Context window

tokens it reads at once

## Where it sits · on the frontier

_Chart: Model Index score of every scored model since late 2022, with the best closed and best open model over time_

o3 reached this score 14 months earlier.

## Test results · by area · tick = best by any model

Show 8 anchor tests

Reasoning & knowledge

evidence

GPQA Diamond

85%

Humanity's Last Exam

no result

Maths

evidence

FrontierMath T1-3

no result

Coding

evidence

SWE-bench Verified

no result

Terminal-Bench

no result

Agents

evidence

APEX-Agents

no result

Novel problems

no evidence yet

ARC-AGI-2

no result

Long tasks

no evidence yet

METR time horizon

no result

Independent results from the Epoch AI benchmarking hub, best reasoning setting. The core tests are shown here; anchor tests also inform the score and can be shown above.

## What others say · rank on each leaderboard

The Model Index

independent tests, our method

#73

of 219 · 145

Artificial Analysis Intelligence Index

overall capability across ten evaluations

#47

of 207 · 23

Epoch Capabilities Index

overall capability from many benchmarks

#71

of 239 · 146.2

Further left is better. Leaderboards measure different things: LMArena is people voting blind between two answers, the others are test scores. Click any row for the source.

## Best at · top three among current models

Not in the top three among current models on any category we track.

Where this model places in the top three of today's models, on category leaderboards and on our core tests.

## Where to use it · prices checked 1 Oct 2026

#### API providers · 3 · per million tokens, in / out

D

DeepInfra

$0.50 / $2.2 · fp4

cheapest

B

BaseTen

$0.60 / $2.4 · fp4

↗

V

Venice

$0.63 / $3.13 · fp8

↗

#### Run it yourself

🤗

Official weights

nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16

↗

Provider prices come from OpenRouter's live listing; going direct to a provider can differ. Subscriptions are the lab's own apps.

## Facts and sources

Released

4 Jun 2026

Announcement

[developer.nvidia.com ↗](https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/)

Release entry

Nemotron 3 Ultra

Scored as

Nemotron 3 Ultra

Artificial Analysis index

23

Output speed

158 tokens/s

Hugging Face

[nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 ↗](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16)

## NVIDIA releases · around this one

Nemotron 3 Nano

Open

15 DEC 2025

Nemotron 3 Super

Open

11 MAR 2026

~132 est.

Nemotron 3 Ultra

Open

4 JUN 2026

index 145

---
Source: https://themodelindex.org/models/nemotron-3-ultra/ · The Model Index · data as of 2026-10-01