Inference
EN

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

Change models
Nemotron 3.5 Super VL 120B A12BNVIDIA
Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIA
IndexAll categories, out of 100.
51.0 ±5.1 best
50.8 ±5.3
GapIndex points below the top.
Top
−0.2 · no clear gap
CoverageIndex weight that has results.
85%
85%
SpeedOutput tokens per second.
—
—
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2026-10
2026-10
Input / 1MUS dollars per 1M input tokens.
—
—
Output / 1MUS dollars per 1M output tokens.
—
—
1K chats2K tokens in, 500 out, each.
—
—
1K documents20K tokens in, 1K out, each.
—
—
1K agent tasks200K tokens in, 20K out, each.
—
—
ContextMaximum tokens in one request.
1M
1M
AgenticMulti-step tasks with tools.
47.8 best
45.1
CodingCode writing and repair.
53.0
55.1 best
ReasoningLogic problems and puzzles.
56.1 best
55.6
MultimodalTasks with images and text.
47.4 best
47.1
KnowledgeFacts and expert knowledge.
52.7
52.7
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
55.2
55.8 best
MathMath problems.
—
—

Benchmarks

23 shared
Only Nemotron 3.5 Super VL 120B A12B: 5
Nemotron 3.5 Super VL 120B A12B
Nemotron 3.5 Super VL 120B A12B NVFP4
Artificial Analysis Humanity's Last ExamKnowledge
30.7%74 / 222
30.7%74 / 222
Artificial Analysis Long Context ReasoningReasoning
68.6% best121 / 217
67.7%127 / 217
Critical Physics TasksReasoning
10.9% best62 / 214
10.6%63 / 214
Artificial Analysis Omniscience AccuracyKnowledge
26.2% best97 / 164
25.9%100 / 164
GDPval-AA normalizedAgentic
33.1%55 / 133
33.1%55 / 133
Artificial Analysis SciCodeCoding
47.8%61 / 114
47.8%61 / 114
Software Engineering Benchmark VerifiedCoding
71.1%64 / 114
71.1%64 / 114
Terminal-Bench 2.1 · BenchLMAgentic
66.7% best56 / 86
60.8%64 / 86
GPQA Diamond · BenchLMKnowledge
89.5%33 / 84
89.9% best31 / 84
SWE-bench ProCoding
51.8% best65 / 83
51.3%67 / 83
Humanity's Last ExamKnowledge
29.2% best49 / 77
29.0%50 / 77
BrowseCompAgentic
78.0% best31 / 63
77.5%32 / 63
LiveCodeBench v6Coding
88.2% best13 / 59
87.4%16 / 59
Instruction Following BenchmarkInstruction
70.9%40 / 57
71.5% best39 / 57
Humanity's Last Exam without toolsKnowledge
31.8% best35 / 53
31.4%38 / 53
CharXiv ReasoningMultimodal
80.8% best31 / 49
79.9%35 / 49
τ³-Bench Tool-Agent-User EvaluationAgentic
18.9% best26 / 32
17.8%27 / 32
MathVisionMultimodal
78.2% best21 / 24
77.8%22 / 24
V*Multimodal
85.6%12 / 14
86.0% best11 / 14
BabyVisionMultimodal
29.6% best10 / 11
29.5%11 / 11