VerdictPal
Join

Models / NVIDIA

Nemotron 3 Ultra 550B A55B (Reasoning)

Nemotron · Released 2026-06-04

NVIDIA's open reasoning MoE, evaluated by Artificial Analysis on 4 Jun. Public catalogs describe 550B total / 55B active parameters and a 262K context window; VerdictPal keeps score claims in the benchmark ledger.

Open reasoning and orchestration model for teams already in NVIDIA's inference stack or testing large open MoEs.

#72 of 168 on AA Index · current snapshot
38 AA Index · 10 benchmark rows · 1 sourcesPosition is benchmark-specific — not a cross-family or cross-source ranking.
Composite52
Context262K
Input / 1M—
Output / 1M—
Knowledge cutoffopen
Output speed87 t/s
TTFT (default API)2.23s
Statuslightweight

Family profile

Best published score in each covered benchmark family.

63/100 avg
Reasoning
87
Coding
40
Agentic
54
Long context
71

4 tested benchmark families

Benchmark placements

Where Nemotron 3 Ultra 550B A55B (Reasoning) places on each public benchmark source that publishes a row. Rank counts every model with a latest row in the same test — not a universal quality score.

BenchmarkFamilyRankScoreSourceDate
IFBenchReasoning#4/ 12781Artificial Analysis2026-09-01
AA time to first tokenPerformance#31/ 186Artificial Analysis2026-09-01
AA output speedPerformance#46/ 187Artificial Analysis2026-09-01
τ³-Bench BankingAgentic#76/ 10614Artificial Analysis2026-09-01
GPQA DiamondReasoning#78/ 18587Artificial Analysis2026-09-01
AA-LCRLong context#80/ 17971Artificial Analysis2026-09-01
Terminal-BenchAgentic#81/ 18454Artificial Analysis2026-09-01
Artificial Analysis Intelligence IndexReasoning#85/ 18738Artificial Analysis2026-09-01
Humanity's Last ExamReasoning#85/ 18028Artificial Analysis2026-09-01
SciCodeCoding#126/ 18040Artificial Analysis2026-09-01
Family context

The three highest-scoring models with pages in each capability family. Where Nemotron 3 Ultra 550B A55B (Reasoning) shows up, it's highlighted.

ReasoningNemotron 3 Ultra 550B A55B (Reasoning) · 87
  1. 1
    Claude Mythos Preview
    95
  2. 2
    Gemini 3.7 Flash
    95
  3. 3
    Grok 4.6
    95
CodingNemotron 3 Ultra 550B A55B (Reasoning) · 40
  1. 1
    GPT 5.5
    96
  2. 2
    Claude Opus 4.8
    96
  3. 3
    Claude Opus 4.6
    96
AgenticNemotron 3 Ultra 550B A55B (Reasoning) · 54
  1. 1
    GPT 5.5
    100
  2. 2
    Claude Opus 4.8
    100
  3. 3
    Claude Fable 5
    100
Long contextNemotron 3 Ultra 550B A55B (Reasoning) · 71
  1. 1
    Muse Spark 1.2
    83
  2. 2
    Kimi K3
    83
  3. 3
    Gemini 3.5 Flash
    81
Newest receipts

Benchmark rows added to the public ledger for Nemotron 3 Ultra 550B A55B (Reasoning) in the last 120 days. Older rows live in the full table below.

10 recent
BenchmarkFamilyScoreSourceDays ago
AA-LCRLong context71%Artificial Analysis5
AA output speedPerformance87 t/sArtificial Analysis5
AA time to first tokenPerformance2.23sArtificial Analysis5
Artificial Analysis Intelligence IndexReasoning38.3%Artificial Analysis5
GPQA DiamondReasoning86.7%Artificial Analysis5
Humanity's Last ExamReasoning28.4%Artificial Analysis5
IFBenchReasoning81.36%Artificial Analysis5
SciCodeCoding39.9%Artificial Analysis5
τ³-Bench BankingAgentic14.23%Artificial Analysis5
Terminal-BenchAgentic53.93%Artificial Analysis5
Family coverage

A high composite that hides a weak family is a trap. These bars surface the families where this model hasn't been publicly tested, and where it leads.

Reasoning87
Coding40
Agentic54
Long context71
Price vs. performance
0255075100$0.10$0.30$1$3$10$30input USD / 1M tokens · log scale →← intelligence

Hover any dot for name, score, and input price. Keyboard: tab through the top twelve, or use the ranking below.

AnthropicOpenAIMistralAlibabaDeepSeekGoogleMoonshotMetaMiniMaxxAIZhipuXiaomiOtherMeituanCohereNVIDIA

Full benchmark ledger

Every catalog benchmark for Nemotron 3 Ultra 550B A55B (Reasoning). Scores link to the original source; gaps mean no public row exists yet.

10 sourced rows

10 of 37 catalog benchmarks have a sourced row for Nemotron 3 Ultra 550B A55B (Reasoning).

Reasoning

Coding

Agentic

Long context

Performance

Sources