Inference
EN

Nemotron 3.5 Super VL 120B A12B NVFP4 vs Qwen3.6-35B-A3B

Change models
Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIA
Qwen3.6-35B-A3BAlibaba
IndexAll categories, out of 100.
50.8 ±5.3 best
50.6 ±3.1
GapIndex points below the top.
Top
−0.2 · no clear gap
CoverageIndex weight that has results.
85%
95% best
SpeedOutput tokens per second.
—
65/s
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2026-10
2026-04
Input / 1MUS dollars per 1M input tokens.
—
$0.10
Output / 1MUS dollars per 1M output tokens.
—
$0.90
1K chats2K tokens in, 500 out, each.
—
$0.65
1K documents20K tokens in, 1K out, each.
—
$2.90
1K agent tasks200K tokens in, 20K out, each.
—
$38.00
ContextMaximum tokens in one request.
1M best
262K
AgenticMulti-step tasks with tools.
45.1
52.7 best
CodingCode writing and repair.
55.1 best
49.6
ReasoningLogic problems and puzzles.
55.6 best
49.9
MultimodalTasks with images and text.
47.1
51.5 best
KnowledgeFacts and expert knowledge.
52.7 best
49.8
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
55.8 best
49.2
MathMath problems.
—
49.0

Benchmarks

11 shared
Only Nemotron 3.5 Super VL 120B A12B NVFP4: 12Only Qwen3.6-35B-A3B: 42
Nemotron 3.5 Super VL 120B A12B NVFP4
Qwen3.6-35B-A3B
Artificial Analysis Humanity's Last ExamKnowledge
30.7% best74 / 222
22.2%106 / 222
Artificial Analysis Long Context ReasoningReasoning
67.7%127 / 217
71.7% best106 / 217
Critical Physics TasksReasoning
10.6% best63 / 214
0.3%137 / 214
GDPval-AA normalizedAgentic
33.1% best55 / 133
19.6%82 / 133
Artificial Analysis SciCodeCoding
47.8% best61 / 114
36.6%98 / 114
Software Engineering Benchmark VerifiedCoding
71.1%64 / 114
73.4% best53 / 114
SWE-bench ProCoding
51.3% best67 / 83
49.5%72 / 83
Humanity's Last ExamKnowledge
29.0% best50 / 77
21.4%62 / 77
LiveCodeBench v6Coding
87.4% best16 / 59
80.4%29 / 59
CharXiv ReasoningMultimodal
79.9% best35 / 49
78.0%39 / 49
τ³-Bench Tool-Agent-User EvaluationAgentic
17.8%27 / 32
67.2% best11 / 32