Inference
EN

Nemotron 3.5 Bolt 74B-A9B vs Nemotron 3.5 Super VL 120B A12B NVFP4

Change models
Nemotron 3.5 Bolt 74B-A9BNVIDIA
Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIA
IndexAll categories, out of 100.
46.4 ±7.9
50.8 ±5.3 best
GapIndex points below the top.
−4.4 · no clear gap
Top
CoverageIndex weight that has results.
70%
85% best
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.
256K
1M best
AgenticMulti-step tasks with tools.
42.3
45.1 best
CodingCode writing and repair.
47.6
55.1 best
ReasoningLogic problems and puzzles.
50.1
55.6 best
MultimodalTasks with images and text.
—
47.1
KnowledgeFacts and expert knowledge.
53.0 best
52.7
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
55.8
MathMath problems.
—
—

Benchmarks

10 shared
Only Nemotron 3.5 Bolt 74B-A9B: 3Only Nemotron 3.5 Super VL 120B A12B NVFP4: 13
Nemotron 3.5 Bolt 74B-A9B
Nemotron 3.5 Super VL 120B A12B NVFP4
Artificial Analysis Humanity's Last ExamKnowledge
24.7%100 / 222
30.7% best74 / 222
Artificial Analysis Long Context ReasoningReasoning
69.7% best113 / 217
67.7%127 / 217
Critical Physics TasksReasoning
4.3%91 / 214
10.6% best63 / 214
GDPval-AA normalizedAgentic
24.0%77 / 133
33.1% best55 / 133
Artificial Analysis SciCodeCoding
39.7%89 / 114
47.8% best61 / 114
Terminal-Bench 2.1 · BenchLMAgentic
51.0%74 / 86
60.8% best64 / 86
GPQA Diamond · BenchLMKnowledge
86.4%51 / 84
89.9% best31 / 84
BrowseCompAgentic
66.8%46 / 63
77.5% best32 / 63
Humanity's Last Exam without toolsKnowledge
25.3%46 / 53
31.4% best38 / 53
τ³-Bench Tool-Agent-User EvaluationAgentic
17.0%28 / 32
17.8% best27 / 32