All models
NVIDIA·2026-06-04·28 evaluations

Nemotron 3 Ultra

Nemotron 3 Ultra 550B A55B (Reasoning)

Open weightsReasoningOpenMDW

Intelligence

23.41

#166 of 520 models

Output speed

161.8 tok/s

Measured generation rate

Cost / index task

$0.579

Measured evaluation workload

End-to-end latency

20ms

First answer 16ms

01 / Position

A model in the landscape

Compare this model with peers across intelligence, speed, cost, and individual evaluations.

Model comparison
Intelligence vs. Time per task
Explore the trade-off: intelligence on the vertical axis, with peers colored by developer.
Selected · Nemotron 3 Ultra26 models plottedBest trade-offs shaded
Most attractive regionLeast attractive regionPareto frontierAnthropicOpenAIMetaNVIDIA
1.1 min4.8 min8.4 min12.1 min15.7 min20.730.640.550.460.4Time per task (minutes)Intelligence IndexClaude Opus 5.5 (max with fallback) · Anthropic Time per task: 13.4 min Intelligence Index: 57.6Claude Opus 5.5 (xhigh with fallback) · Anthropic Time per task: 8.4 min Intelligence Index: 56.0Claude Sonnet 5.5 (max with fallback) · Anthropic Time per task: 14.5 min Intelligence Index: 56.0Claude Opus 5.5 (high with fallback) · Anthropic Time per task: 4.9 min Intelligence Index: 53.6Claude Fable 5.1 (max with fallback) · Anthropic Time per task: 12.1 min Intelligence Index: 53.4Claude Fable 5.1 (xhigh with fallback) · Anthropic Time per task: 11.3 min Intelligence Index: 53.2GPT-6 Astra (max) · OpenAI Time per task: 7.8 min Intelligence Index: 52.8GPT-6 Astra (xhigh) · OpenAI Time per task: 5.5 min Intelligence Index: 52.5Claude Sonnet 5.5 (xhigh with fallback) · Anthropic Time per task: 7.4 min Intelligence Index: 51.9GPT-6.1 Sol (max) · OpenAI Time per task: 9.5 min Intelligence Index: 51.8Claude Opus 5.5 (medium with fallback) · Anthropic Time per task: 3.6 min Intelligence Index: 51.2Claude Fable 5.1 (high with fallback) · Anthropic Time per task: 7.4 min Intelligence Index: 51.2GPT-6 Astra (high) · OpenAI Time per task: 4.0 min Intelligence Index: 51.0GPT-6.1 Sol (xhigh) · OpenAI Time per task: 4.5 min Intelligence Index: 51.0Claude Opus 5 (max) · Anthropic Time per task: 14.7 min Intelligence Index: 50.7GPT-6.1 Sol (high) · OpenAI Time per task: 3.4 min Intelligence Index: 50.2Claude Fable 5 (with fallback) · Anthropic Time per task: 11.4 min Intelligence Index: 49.7GPT-6 Astra (medium) · OpenAI Time per task: 3.3 min Intelligence Index: 49.7Claude Opus 5 (xhigh) · Anthropic Time per task: 12.2 min Intelligence Index: 49.7Claude Fable 5.1 (medium with fallback) · Anthropic Time per task: 5.6 min Intelligence Index: 49.1Claude Opus 5 (high) · Anthropic Time per task: 9.1 min Intelligence Index: 48.2Muse Spark 1.3 (max) · Meta Time per task: 5.4 min Intelligence Index: 48.2GPT-6.1 Sol (medium) · OpenAI Time per task: 2.1 min Intelligence Index: 47.8GPT-6 Sol (max) · OpenAI Time per task: 6.8 min Intelligence Index: 47.5GPT-5.6 Sol (max) · OpenAI Time per task: 8.0 min Intelligence Index: 47.1Nemotron 3 Ultra · NVIDIA Time per task: 2.7 min Intelligence Index: 23.4 Current modelNemotron 3 Ultra
26 of 26 models with both measuresNemotron 3 Ultra: 2.7 min · 23.4

02 / Evidence

Benchmark results

28 distinct evaluations grouped by topic. Explore cross-model charts or switch to the full results list.

Explore all models
Benchmark landscape
AA-Briefcase across models
Compare evaluation scores against operating cost, latency, or throughput.
Selected · Nemotron 3 Ultra190 scored models878.22 · #119

Model positions

AA-Briefcase · highest score first · 190 models

Nemotron 3 Ultra is #119 of 190 · 878.2
Most attractive regionLeast attractive regionPareto frontierAnthropicSpaceXAIOpenAIAlibabaMetaNVIDIA
$0.00$2.36$4.72$7.08$9.450%491.9983.81,475.71,967.6Cost per task (USD)AA-BriefcaseClaude Opus 5.5 (max with fallback) · Anthropic Cost per task: $5.98 AA-Briefcase: 1,821.9Claude Sonnet 5.5 (max with fallback) · Anthropic Cost per task: $7.60 AA-Briefcase: 1,810.9Claude Opus 5.5 (xhigh with fallback) · Anthropic Cost per task: $3.46 AA-Briefcase: 1,779.5Claude Sonnet 5.5 (xhigh with fallback) · Anthropic Cost per task: $2.74 AA-Briefcase: 1,745.7Claude Opus 5.5 (high with fallback) · Anthropic Cost per task: $1.82 AA-Briefcase: 1,704.5Grok 4.7 (xhigh) · SpaceXAI Cost per task: $3.74 AA-Briefcase: 1,657.2Claude Fable 5.1 (xhigh with fallback) · Anthropic Cost per task: $5.98 AA-Briefcase: 1,650Claude Fable 5.1 (max with fallback) · Anthropic Cost per task: $7.63 AA-Briefcase: 1,646.1Claude Opus 5.5 (medium with fallback) · Anthropic Cost per task: $1.34 AA-Briefcase: 1,642.1Grok 4.7 (high) · SpaceXAI Cost per task: $2.73 AA-Briefcase: 1,636.7Claude Sonnet 5.5 (high with fallback) · Anthropic Cost per task: $1.08 AA-Briefcase: 1,634Claude Opus 5 (max) · Anthropic Cost per task: $5.86 AA-Briefcase: 1,633.4Claude Opus 5 (xhigh) · Anthropic Cost per task: $4.88 AA-Briefcase: 1,624.6Qwen3.8 Max (0902) · Alibaba Cost per task: $5.41 AA-Briefcase: 1,609.7Muse Spark 1.3 (max) · Meta Cost per task: $1.60 AA-Briefcase: 1,589.2Qwen3.8-Flash-Next · Alibaba Cost per task: $0.37 AA-Briefcase: 1,586.6Claude Fable 5.1 (high with fallback) · Anthropic Cost per task: $3.91 AA-Briefcase: 1,578.2GPT-6.1 Sol (max) · OpenAI Cost per task: $0.72 AA-Briefcase: 1,564.2GPT-6 Astra (max) · OpenAI Cost per task: $3.26 AA-Briefcase: 1,562Claude Opus 5 (high) · Anthropic Cost per task: $3.61 AA-Briefcase: 1,556.7Grok 4.6 (xhigh) · SpaceXAI Cost per task: $2.32 AA-Briefcase: 1,545.1Grok 4.6 (high) · SpaceXAI Cost per task: $1.86 AA-Briefcase: 1,534.1GPT-6 Astra (xhigh) · OpenAI Cost per task: $2.31 AA-Briefcase: 1,533.7Claude Fable 5.1 (medium with fallback) · Anthropic Cost per task: $2.98 AA-Briefcase: 1,531.2Claude Fable 5 (with fallback) · Anthropic Cost per task: $8.75 AA-Briefcase: 1,529.6Nemotron 3 Ultra · NVIDIA Cost per task: $0.58 AA-Briefcase: 878.2 Current modelNemotron 3 Ultra
26 of 180 models with both measuresNemotron 3 Ultra: $0.58 · 878.2

Domain capability

Finance And Accounting24.65
Strategy And Ops17.19
Legal27.04
Healthcare And Medical23.6
Engineering27.66
Economics37.2

03 / Operation

Speed, tokens & price

Prices are per million tokens; latency and task costs describe the captured evaluation rather than a universal provider guarantee.

Token pricing

Input / 1M
$0.600
Output / 1M
$2.40
Cache read / 1M
$0.200
Cache write / 1M
—
Cache-hit discount
80%

Response & usage

Output speed
161.8 tok/s
First chunk
2ms
First answer
16ms
Output / task
28.4k
Reasoning / task
15.1k

Evaluation workload

Canonical input
1.07B
Canonical output
112.45M
Answer / task
13.3k
Index time / task
2.7m
Run cost
$778.138

By prompt type

PromptOutput speedFirst chunkFirst answerEnd to end
Medium156.3 tok/s1ms16ms19ms
Long161.8 tok/s2ms16ms20ms
Hundred K197.9 tok/s5ms17ms19ms
Medium Parallel139.3 tok/s2ms18ms22ms

04 / Access

License & availability

Recorded license

OpenMDW

The publisher-linked license text defines permitted use and redistribution. Review it before using the model.

Weight access
Open weights
Classification
Permissive
Reasoning
Yes

Routing and providers

OpenRouter ID
nvidia/nemotron-3-ultra-550b-a55b
Tokenizer
Other
Instruct type
—
Captured endpoints
4
ProviderQuantizationContexttok/s
DeepInfrafp4262.1k—
BaseTenfp4202.8k—
BaseTenfp4202.8k—
Venicefp8256k—

05 / Audit

Underlying fields

Inspect the captured model and routing records. Artificial Analysis destination links are omitted.