ZMIME
NVIDIA · Released Jun 4, 2026

Nemotron 3 Ultra 550B A55B

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

  • Open weights
  • Reasoning
  • Tool calling
Capability
147.0ECI · #82 of 158
Input
$0.55per 1M tokens
Output
$2.40per 1M tokens
Context
1M128K max output
Pricing

What it costs

Input
$0.55per million tokens
Output
$2.40per million tokens
Blended (3:1)
$1.01pricier than 59% of priced models

Typical monthly bills

Monthly usageEstimated cost
Side project2M input + 0.5M output tokens$2.30
Team assistant25M input + 5M output tokens$25.75
Production app250M input + 50M output tokens$257.50

Median of 20 paid API providers on models.dev; NVIDIA also offers it free on Nvidia.

Capability

Independent benchmarks

147.0Epoch Capabilities Index
#82 of 158 scored models
68Median 147174

The shaded band is Epoch AI’s confidence range (145.0–149.1); the tick marks the median scored model.

  • GPQA DiamondGraduate-level science questions85.4%
  • OTIS Mock AIME 2024–2025Competition mathematics86.7%

Scores from Epoch AI, run independently of NVIDIA.

Specifications

The details

Lab
NVIDIA
Released
Jun 4, 2026
Knowledge cutoff
Sep 2025
Context window
1,000,000 tokens
Max output
128,000 tokens
Inputs
Text
Output
Text
Reasoning
Adjustable effort medium · high
Tool calling
Yes
Structured output
Yes
Weights
Open weights
API model ID
nvidia/nemotron-3-ultra-550b-a55bon Nvidia
Availability
26 API providerslisted on models.dev

Nvidia model documentation

Lab-reported

What NVIDIA claims

Published by the lab at launch. Settings vary, so compare these only with care.

BenchmarkScoreSettingSource
SWE-Bench Verified70.7 resolved—Source
SWE-Bench Multilingual67.7 resolve rate—Source
Terminal-Bench v2.156.4 success rate—Source
GPQA87no toolsSource
Humanity's Last Exam26.7no toolsSource
Humanity's Last Exam37.4with toolsSource
LiveCodeBench vv689—Source
MMLU-Pro86.8—Source
BrowseComp44.4—Source
IFBench81.7prompt looseSource
GDPval46.7 wins or ties—Source
Same lab

More from NVIDIA

Questions

About Nemotron 3 Ultra 550B A55B

How much does Nemotron 3 Ultra 550B A55B cost?

Nemotron 3 Ultra 550B A55B costs $0.55 input / $2.40 output per million tokens (median across 20 API providers; free on Nvidia). At a 3:1 input-to-output mix that is $1.01 per million tokens, more expensive than 59% of the 469 priced models we track.

What is the context window of Nemotron 3 Ultra 550B A55B?

Nemotron 3 Ultra 550B A55B accepts up to 1,000,000 tokens per request and can write up to 128,000 tokens in one response.

How good is Nemotron 3 Ultra 550B A55B?

Epoch AI gives Nemotron 3 Ultra 550B A55B a Capabilities Index score of 147.0 (likely range 145.0–149.1), ranking it #82 of 158 models Epoch has scored. Epoch AI benchmark results: GPQA Diamond 85.4%, OTIS Mock AIME 2024–2025 86.7%.

Is Nemotron 3 Ultra 550B A55B open source?

Yes. NVIDIA publishes the weights, so it can be downloaded and self-hosted.

What inputs does Nemotron 3 Ultra 550B A55B support?

Nemotron 3 Ultra 550B A55B accepts text and replies in text. It is a reasoning model with medium and high effort settings, supports tool calling and can return structured JSON output.

When was Nemotron 3 Ultra 550B A55B released?

NVIDIA released Nemotron 3 Ultra 550B A55B on Jun 4, 2026. Its training data runs to Sep 2025.

What are the best alternatives to Nemotron 3 Ultra 550B A55B?

The closest current models from other labs on capability, price and release date are Qwen3.6 27B (Alibaba (Qwen), ECI 146.8, $0.60 / $3.60), Gemini 3.5 Flash Lite (Google, ECI 145.7, $0.30 / $2.50), MiniMax-M3 (MiniMax, ECI 147.9, $0.30 / $1.20) and GLM-5 (Z.ai (Zhipu), ECI 146.5, $1.00 / $3.20).