Inference
EN

Nemotron 3.5 Super VL 120B A12B NVFP4 vs Nemotron 3 Ultra

Change models
Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIA
Nemotron 3 UltraNVIDIA
IndexAll categories, out of 100.
50.8 ±5.3
50.8 ±5.3
GapIndex points below the top.
Top
Top
CoverageIndex weight that has results.
85% best
80%
SpeedOutput tokens per second.
—
—
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2026-10
2026-06
Input / 1MUS dollars per 1M input tokens.
—
Free
Output / 1MUS dollars per 1M output tokens.
—
Free
1K chats2K tokens in, 500 out, each.
—
Free
1K documents20K tokens in, 1K out, each.
—
Free
1K agent tasks200K tokens in, 20K out, each.
—
Free
ContextMaximum tokens in one request.
1M
1M
AgenticMulti-step tasks with tools.
45.1
48.9 best
CodingCode writing and repair.
55.1 best
54.3
ReasoningLogic problems and puzzles.
55.6 best
50.0
MultimodalTasks with images and text.
47.1
—
KnowledgeFacts and expert knowledge.
52.7 best
50.0
MultilingualTasks in many languages.
—
52.4
InstructionTasks with strict prompt rules.
55.8
65.2 best
MathMath problems.
—
—

Benchmarks

15 shared
Only Nemotron 3.5 Super VL 120B A12B NVFP4: 8Only Nemotron 3 Ultra: 17
Nemotron 3.5 Super VL 120B A12B NVFP4
Nemotron 3 Ultra
Artificial Analysis Humanity's Last ExamKnowledge
30.7% best74 / 222
28.4%89 / 222
Artificial Analysis Long Context ReasoningReasoning
67.7% best127 / 217
67.0%130 / 217
Critical Physics TasksReasoning
10.6% best63 / 214
3.1%95 / 214
Artificial Analysis Omniscience AccuracyKnowledge
25.9% best100 / 164
21.6%123 / 164
GDPval-AA normalizedAgentic
33.1%55 / 133
33.1%55 / 133
Artificial Analysis SciCodeCoding
47.8% best61 / 114
40.3%86 / 114
Software Engineering Benchmark VerifiedCoding
71.1%64 / 114
71.9% best63 / 114
Terminal-Bench 2.1 · BenchLMAgentic
60.8% best64 / 86
56.4%68 / 86
GPQA Diamond · BenchLMKnowledge
89.9% best31 / 84
87.0%49 / 84
Humanity's Last ExamKnowledge
29.0% best50 / 77
26.7%55 / 77
BrowseCompAgentic
77.5% best32 / 63
44.4%61 / 63
LiveCodeBench v6Coding
87.4%16 / 59
89.0% best12 / 59
Instruction Following BenchmarkInstruction
71.5%39 / 57
81.7% best5 / 57
Humanity's Last Exam without toolsKnowledge
31.4% best38 / 53
26.7%44 / 53
τ³-Bench Tool-Agent-User EvaluationAgentic
17.8%27 / 32
70.9% best4 / 32