Nemotron 3 Ultra

Tool callingStructured outputReasoning

Hugging Face Launch post Website

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

Context: 262K
Released: 4mo ago
Parameters: 550B
Input / 1M tokens: $0.60
Output / 1M tokens: $2.40
Cached input / 1M tokens: $0.12

Sources:models.devOpenRouter

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.

Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.

Other published benchmarks

Evaluation	Reported score and source
GPQA diamond	85.4% Epoch AI · nemotron-3-ultra Epoch AI evaluation
OTIS Mock AIME 2024-2025	86.7% Epoch AI · nemotron-3-ultra Epoch AI evaluation

Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.

Specifications

Developer: NVIDIA
Model family: Nemotron
Type: Language model
Predecessor: —
Successor: —
Released: June 4, 2026
Model ID: nvidia/nemotron-3-ultra-550b-a55b
Openness: Open weights
License: —
Parameters: 550B
Input modalities: text
Output modalities: text
Modality: text->text
Native context: 262K
Tokenizer: Other
Max output tokens: 183K
Hugging Face: nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
Source freshness: Metadata synced Oct 8, 2026, 10:17 PM UTC

Additional source evidence

Source model: nvidia/nemotron-3-ultra-550b-a55b
Source family: nemotron
Last updated: 2026-06-04
Model context: 1M
Model output limit: 128K
Open weights: Yes
File attachments: No
Reasoning: Yes
Tool calling: Yes
Temperature control: Yes

Published benchmarks

Benchmark	Score	Evaluation details	Evidence
SWE-Bench Verified	70.7 · resolved	2026-06-04	Source
SWE-Bench Multilingual	67.7 · resolve rate	2026-06-04	Source
Terminal-Bench	56.4 · success rate	2.1 · 2026-06-04	Source
GPQA	87 · accuracy	no tools · 2026-06-04	Source
Humanity's Last Exam	26.7 · accuracy	no tools · 2026-06-04	Source
Humanity's Last Exam	37.4 · accuracy	with tools · 2026-06-04	Source
LiveCodeBench	89 · pass@1	v6 · 2026-06-04	Source
MMLU-Pro	86.8 · accuracy	2026-06-04	Source
BrowseComp	44.4 · accuracy	2026-06-04	Source
IFBench	81.7 · accuracy	prompt loose · 2026-06-04	Source
GDPval	46.7 · wins or ties	2026-06-04	Source

Complete source record (JSON)

{
  "id": "nvidia/nemotron-3-ultra-550b-a55b",
  "name": "Nemotron 3 Ultra 550B A55B",
  "description": "Largest Nemotron 3 model for maximum open-weight reasoning and agent accuracy",
  "family": "nemotron",
  "last_updated": "2026-06-04",
  "attachment": false,
  "reasoning": true,
  "tool_call": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 1000000,
    "output": 128000
  },
  "benchmarks": [
    {
      "name": "SWE-Bench Verified",
      "score": 70.7,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "resolved",
      "date": "2026-06-04"
    },
    {
      "name": "SWE-Bench Multilingual",
      "score": 67.7,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "resolve rate",
      "date": "2026-06-04"
    },
    {
      "name": "Terminal-Bench",
      "score": 56.4,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "success rate",
      "version": "2.1",
      "date": "2026-06-04"
    },
    {
      "name": "GPQA",
      "score": 87,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "accuracy",
      "variant": "no tools",
      "date": "2026-06-04"
    },
    {
      "name": "Humanity's Last Exam",
      "score": 26.7,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "accuracy",
      "variant": "no tools",
      "date": "2026-06-04"
    },
    {
      "name": "Humanity's Last Exam",
      "score": 37.4,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "accuracy",
      "variant": "with tools",
      "date": "2026-06-04"
    },
    {
      "name": "LiveCodeBench",
      "score": 89,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "pass@1",
      "version": "v6",
      "date": "2026-06-04"
    },
    {
      "name": "MMLU-Pro",
      "score": 86.8,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "accuracy",
      "date": "2026-06-04"
    },
    {
      "name": "BrowseComp",
      "score": 44.4,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "accuracy",
      "date": "2026-06-04"
    },
    {
      "name": "IFBench",
      "score": 81.7,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "accuracy",
      "variant": "prompt loose",
      "date": "2026-06-04"
    },
    {
      "name": "GDPval",
      "score": 46.7,
      "source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "metric": "wins or ties",
      "date": "2026-06-04"
    }
  ],
  "release_date": "2026-06-04",
  "open_weights": true
}

Infrastructure

Provider availability

All providers

Provider	Endpoint	Context	Input /M tokens	Output /M tokens	Cached input /M tokens	Cache write /M tokens	Uptime 1d	Data handling
DeepInfra	deepinfra/fp4	262K	$0.50	$2.20	$0.10	—	100.00%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
BaseTen	baseten/fp4	203K	$0.60	$2.40	$0.12	—	100.00%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
BaseTen	baseten/fp4/us	203K	$0.60	$2.40	$0.12	—	99.92%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
Venice	venice/fp8	256K	$0.63	$3.13	$0.19	—	99.68%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15

Provider data:OpenRouter

Deployment graph

Can I run this?

Vertical Tabs only shows compatibility when it has structured evidence tying this exact model to hardware, runtime, quantization, memory use, and observed performance.

Browse hardware