‹ PublicAI Index
The LLM benchmark aggregator.
Nemotron 3 Ultra 550B A55B
NVIDIA · 550B · A55B
Strongest in Instruction following (#14 of 57), weakest in Safety (#256 of 337). Above par in 5 of 19 scopes. Among the models it meets almost everywhere, it finishes behind Grok 4.7 and GPT-6 Astra and ahead of Grok Build 0.1 and Command A.
§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Agents52.6−15.5#91/2682/5
Knowledge work53.4−20.2#61/1782/2
Reasoning44−22.9#134/1781/4
Mathematics49.3−16.4#83/1401/2
Reasoning37.4−31.7#126/1401/3
Coding39.1−30#159/1651/5
Code generation36.3−33.7#67/771/2
Agentic coding40.9−27.8#138/1571/4
Core abilities39.4−28#190/2041/3
Instruction following59.6−14.1#14/571/1
Language34−37.5#54/571/1
Data analysis26−40.1#55/571/1
General intelligence38.1−32#185/2041/3
Safety47−14.5#256/3371/3
Secure code63.5−3#26/2741/1
Toxicity avoidance52.4−6.6#167/2721/1
Fairness43.9−26.1#222/3001/2
Jailbreak resistance32.9−32#246/2721/1
Harm refusal44.9−16.7#250/3001/2
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Nemotron 3 Ultra 550B A55B, left for the other.
§ 3 · Sources
Where the numbers come from
4 publications, 16 figures. Every one links to the page it was read from.
GDPval-AA 1000
AA-Briefcase 876
LiveBench 67.4LiveBench · Reasoning 74.7LiveBench · Coding 70.7LiveBench · Agentic Coding 38.7LiveBench · Mathematics 88.7LiveBench · Data Analysis 54.5LiveBench · Language 70.8LiveBench · Instruction Following 73.4
Enkrypt · Jailbreak risk 27.2%Enkrypt · Harmful content risk 9.4%Enkrypt · CBRN risk 52.7%Enkrypt · Toxicity risk 4.6%Enkrypt · Bias risk 87.3%Enkrypt · Insecure code risk 6.2%
Badge
[](https://publicai.io/model-index/m/nemotron-3-ultra-550b-a55b)