NVIDIA AI & NIMopen-weight
Released: 2025-03-20

Nemotron 3 Super 120B

High-efficiency 120B Mixture-of-Experts with 12B active parameters per token, optimized for TensorRT-LLM deployment.

Intelligence Index
85.0/ 100
Calibrated multi-domain composite
Throughput Speed
130tok/sec
Output streaming throughput
Time To First Token
230ms
Initial chunk server latency
LiveBench Contam-Free
74.2%
September 2026 suite score

Sourced Benchmark Evaluations (61)

Verified performance across authoritative benchmarks with provenance tracking.
BenchmarkStatusScoreTrustDateSource TypeProvenance
ARC-AGI-3Active54.9%
93
2025-04-24independentSource ↗
ARC-AGI-2Nearing Saturation65.9%
75
2025-04-17independentSource ↗
FrontierMathActive23.5%
98
2025-04-14independentSource ↗
PutnamBenchActive43.7%
90
2025-04-13independentSource ↗
SWE-LancerDeprecated53%
20
2025-04-11independentSource ↗
Omni-MATHDeprecated62.2%
10
2025-04-11independentSource ↗
SoohakActive63.8%
98
2025-04-09independentSource ↗
FrontierCodeActive45.2%
98
2025-04-09independentSource ↗
Humanity's Last ExamActive37.2%
93
2025-04-07independentSource ↗
StructFlowBenchNearing Saturation66.5%
75
2025-04-07independentSource ↗
AIMENearing Saturation79.8%
75
2025-04-07independentSource ↗
OlympiadBenchActive68.9%
80
2025-04-06independentSource ↗
MathArenaActive67.2%
99
2025-04-05independentSource ↗
FireBenchActive65.7%
95
2025-04-05independentSource ↗
Terminal-BenchActive54.6%
80
2025-04-05independentSource ↗
OpenAI-MRCRActive71.4%
80
2025-04-04independentSource ↗
SimpleQAActive44.2%
90
2025-04-04independentSource ↗
CodeContestsDeprecated50.7%
10
2025-04-04independentSource ↗
ComplexBenchNearing Saturation64.1%
80
2025-04-03independentSource ↗
SWE-bench ProActive48.4%
90
2025-04-03independentSource ↗
LongBench-ProActive69.7%
80
2025-04-03independentSource ↗
ZeroSCROLLSNearing Saturation68%
78
2025-04-02independentSource ↗
MultiPL-EDeprecated71.8%
10
2025-04-02independentSource ↗
LiveCodeBenchSaturated64%
39
2025-04-01independentSource ↗
AgentIFNearing Saturation53.2%
80
2025-04-01independentSource ↗
MATHSaturated82.3%
25
2025-04-01independentSource ↗
InfiniteBenchNearing Saturation72.3%
65
2025-04-01independentSource ↗
MercuryDeprecated68.6%
10
2025-03-31independentSource ↗
SCROLLSNearing Saturation68%
65
2025-03-31independentSource ↗
MMLU-ProNearing Saturation78.2%
75
2025-03-31independentSource ↗
RULERNearing Saturation78.2%
84
2025-03-30independentSource ↗
TabMWPSaturated77.3%
25
2025-03-30independentSource ↗
CyberSecEvalDeprecated70.2%
10
2025-03-30independentSource ↗
SWE-bench VerifiedSaturated60.8%
25
2025-03-29independentSource ↗
MMLU-ReduxSaturated83.3%
25
2025-03-29independentSource ↗
QuALITYNearing Saturation74.8%
78
2025-03-29independentSource ↗
LongBenchNearing Saturation73.1%
65
2025-03-28independentSource ↗
SWE-benchDeprecated53%
10
2025-03-28independentSource ↗
MGSMNearing Saturation79.8%
65
2025-03-28independentSource ↗
AGIEvalSaturated72.7%
25
2025-03-27independentSource ↗
QASPERNearing Saturation70.6%
65
2025-03-27independentSource ↗
TruthfulQASaturated74.8%
25
2025-03-27independentSource ↗
GSM-HardDeprecated80.6%
10
2025-03-26independentSource ↗
SuperGLUESaturated86.2%
25
2025-03-26independentSource ↗
CodeXGLUEDeprecated73.3%
23
2025-03-25independentSource ↗
MT-BenchNearing Saturation8.1%
75
2025-03-25independentSource ↗
BIG-Bench HardSaturated82.8%
25
2025-03-25independentSource ↗
WinoGrandeSaturated87.9%
28
2025-03-24independentSource ↗
MBPP+Nearing Saturation79.6%
65
2025-03-24independentSource ↗
IFEvalDeprecated76.6%
25
2025-03-24independentSource ↗
GPQANearing Saturation74.8%
78
2025-03-24independentSource ↗
Needle-in-a-HaystackSaturated97%
44
2025-03-23independentSource ↗
HumanEval+Saturated84.2%
25
2025-03-23independentSource ↗
CommonsenseQASaturated88.7%
28
2025-03-23independentSource ↗
MBPPSaturated82.7%
25
2025-03-22independentSource ↗
GSM8KSaturated90.7%
25
2025-03-22independentSource ↗
GLUESaturated91.3%
28
2025-03-22independentSource ↗
MMLUSaturated86.7%
25
2025-03-21independentSource ↗
HumanEvalSaturated88.9%
25
2025-03-21independentSource ↗
ARC-AGISaturated79.4%
25
2025-03-21independentSource ↗
LiveBench (Continuous)Active74.2%
94
2025-02-01independentSource ↗
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run nvidia-nemotron-3-super-120b

# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nvidia-nemotron-3-super-120b",
    "messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
  }'

Evaluation & Sourcing Notes

Scores listed for Nemotron 3 Super 120B represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.

All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.