NVIDIA AI & NIMopen-weight
Released: 2024-06-14

Nemotron-4-340B-Reward

Top-performing reward model on RewardBench for scoring alignment, safety, helpfulness, and response quality in RLHF loops.

Intelligence Index
86.0/ 100
Calibrated multi-domain composite
Throughput Speed
45tok/sec
Output streaming throughput
Time To First Token
480ms
Initial chunk server latency
LiveBench Contam-Free
74.5%
September 2026 suite score

Sourced Benchmark Evaluations (51)

Verified performance across authoritative benchmarks with provenance tracking.
BenchmarkStatusScoreTrustDateSource TypeProvenance
LiveBench (Continuous)Active74.5%
94
2025-02-01independentSource ↗
ARC-AGI-3Active55.3%
93
2024-07-19independentSource ↗
ARC-AGI-2Nearing Saturation66.3%
75
2024-07-12independentSource ↗
FrontierMathActive23.5%
98
2024-07-09independentSource ↗
PutnamBenchActive43.7%
90
2024-07-08independentSource ↗
SWE-LancerDeprecated53%
20
2024-07-06independentSource ↗
Omni-MATHDeprecated62.2%
10
2024-07-06independentSource ↗
SoohakActive63.8%
98
2024-07-04independentSource ↗
FrontierCodeActive45.2%
98
2024-07-04independentSource ↗
AIMENearing Saturation79.8%
75
2024-07-02independentSource ↗
StructFlowBenchNearing Saturation67.3%
75
2024-07-02independentSource ↗
Humanity's Last ExamActive37.4%
93
2024-07-02independentSource ↗
OlympiadBenchActive68.9%
80
2024-07-01independentSource ↗
MathArenaActive67.2%
99
2024-06-30independentSource ↗
Terminal-BenchActive54.6%
80
2024-06-30independentSource ↗
FireBenchActive66.5%
95
2024-06-30independentSource ↗
SimpleQAActive44.7%
90
2024-06-29independentSource ↗
CodeContestsDeprecated50.7%
10
2024-06-29independentSource ↗
ComplexBenchNearing Saturation64.9%
80
2024-06-28independentSource ↗
SWE-bench ProActive48.4%
90
2024-06-28independentSource ↗
MultiPL-EDeprecated71.8%
10
2024-06-27independentSource ↗
LiveCodeBenchSaturated64%
39
2024-06-26independentSource ↗
MATHSaturated82.3%
25
2024-06-26independentSource ↗
AgentIFNearing Saturation53.8%
80
2024-06-26independentSource ↗
MercuryDeprecated68.6%
10
2024-06-25independentSource ↗
MMLU-ProNearing Saturation79.1%
75
2024-06-25independentSource ↗
CyberSecEvalDeprecated70.2%
10
2024-06-24independentSource ↗
TabMWPSaturated77.3%
25
2024-06-24independentSource ↗
SWE-bench VerifiedSaturated60.8%
25
2024-06-23independentSource ↗
MMLU-ReduxSaturated84.3%
25
2024-06-23independentSource ↗
SWE-benchDeprecated53%
10
2024-06-22independentSource ↗
MGSMNearing Saturation79.8%
65
2024-06-22independentSource ↗
TruthfulQASaturated75.7%
25
2024-06-21independentSource ↗
AGIEvalSaturated73.1%
25
2024-06-21independentSource ↗
SuperGLUESaturated86.7%
25
2024-06-20independentSource ↗
GSM-HardDeprecated80.6%
10
2024-06-20independentSource ↗
MT-BenchNearing Saturation8.1%
75
2024-06-19independentSource ↗
BIG-Bench HardSaturated83.3%
25
2024-06-19independentSource ↗
CodeXGLUEDeprecated73.3%
23
2024-06-19independentSource ↗
MBPP+Nearing Saturation79.6%
65
2024-06-18independentSource ↗
IFEvalDeprecated77.5%
25
2024-06-18independentSource ↗
WinoGrandeSaturated88.4%
28
2024-06-18independentSource ↗
GPQANearing Saturation75.7%
78
2024-06-18independentSource ↗
CommonsenseQASaturated89.3%
28
2024-06-17independentSource ↗
HumanEval+Saturated84.2%
25
2024-06-17independentSource ↗
GLUESaturated91.8%
28
2024-06-16independentSource ↗
MBPPSaturated82.7%
25
2024-06-16independentSource ↗
GSM8KSaturated90.7%
25
2024-06-16independentSource ↗
MMLUSaturated87.7%
25
2024-06-15independentSource ↗
HumanEvalSaturated88.9%
25
2024-06-15independentSource ↗
ARC-AGISaturated79.9%
25
2024-06-15independentSource ↗
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run nemotron-4-340b-reward

# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nemotron-4-340b-reward",
    "messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
  }'

Evaluation & Sourcing Notes

Scores listed for Nemotron-4-340B-Reward 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.