Back to nemotron
NVIDIA /

Nemotron 3 Ultra 550B A55B
Nemotron 3 Ultra 550B A55B

Open WeightsChat & ReasoningtextUpdated June 4, 2026
Some details or benchmark scores on this page are self-reported by developers and unconfirmed.

Nemotron 3 Ultra 550B A55B

📌 Model Overview

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

Nemotron 3 Ultra 550B A55B is a Open Weights model developed by NVIDIA, released on 2026-06-04. It is designed primarily for Chat Reasoning workloads. Featuring a 1M tokens context window and 550B parameter count, it offers robust performance for enterprise integration, developers, and researchers.

✨ Key Features & Capabilities

Specification Table
FeatureDescription
Context Window1M tokens capacity for extended prompts and multi-turn workflows
Primary TaskOptimized for Chat Reasoning
Deploymentself-hostable, api-only
Modalitytext
Max OutputMax Output: 128K tokens
Native reasoning capabilityNative reasoning capability
Tool / function calling supportTool / function calling support

⚙️ Technical Specifications

Specification Table
SpecificationDetails
Developer / LabNVIDIA
Release Date2026-06-04
Model TypeOpen Weights
Parameters550B
Context Window1M tokens
Licenseproprietary
Model Familynemotron

📊 Benchmarks & Performance

Specification Table
BenchmarkScoreSource
SWE-Bench Verified70.7%Independent Eval
SWE-Bench Multilingual67.7 (resolve rate)Independent Eval
Terminal-Bench56.4%Independent Eval
GPQA87 (accuracy)Independent Eval
Humanity's Last Exam26.7 (accuracy)Independent Eval
Humanity's Last Exam37.4 (accuracy)Independent Eval
LiveCodeBench89 (pass@1)Independent Eval
MMLU-Pro86.8 (accuracy)Independent Eval
BrowseComp44.4 (accuracy)Independent Eval
IFBench81.7 (accuracy)Independent Eval
GDPval46.7 (wins or ties)Independent Eval

💰 Pricing

Specification Table
Tier / UnitRate (USD)
1M input tokens$0.5
1M output tokens$2.5
1M cache read tokens$0.15

📜 License & Usage

This model is governed by the proprietary license. Please check official developer guidelines before commercial deployment.

Key Features

Max Output: 128K tokens

Feature 01✦

Native reasoning capability

Feature 02✦

Tool / function calling support

Feature 03✦

You might also want to compare

Kimi K3

Moonshot AI

Specs →

Claude Opus 5

Anthropic

Specs →

KAT-Coder-V2.5-Dev

Kwaipilot

Specs →

Verified Sources

Tags

reasoningtool-callingopen-weights

Model Specs

open-weights

Parameters

550B

Context Window

1M tokens

License

proprietary

Deployment

self-hostableapi-only

Pricing

  • 1M input tokens$0.5
  • 1M output tokens$2.5
  • 1M cache read tokens$0.15

as of 2026-07-26

Lineage

Model Family

Part of the nemotron family

Nemotron 3.5 Content SafetyNemotron 3 Nano Omni 30B A3B ReasoningNemotron 3 Content SafetyLlama Nemotron Rerank VL 1B v2Nemotron Cascade 2 30B A3BNemotron VoiceChatNemotron 3 Super 120B A12BLlama Nemotron Embed VL 1B v2Nemotron Content Safety Reasoning 4BNemotron 3 Nano 30B A3BLlama 3.1 Nemotron Safety Guard 8B v3Nemotron Nano 12B v2 VLNemotron Nano 9B v2Llama 3.3 Nemotron Super 49B v1.5Mistral NemotronLlama 3.1 Nemotron 70B InstructLlama 3.1 Nemotron Ultra 253BLlama 3.3 Nemotron Super 49B v1Nemotron Mini 4B Instruct

Curator Notes

Skeleton entry imported from models.dev backfill (2026-07-26). Needs manual enrichment and primary-source verification.

Compare Specs

Compare parameters, context windows, modalities, and benchmark scores of this model side-by-side with others.

Compare Model