docketrouter
Models / NVIDIA

Nemotron 3 Super

by NVIDIA · nvidia/nemotron-3-super-120b-a12b

NVIDIA Nemotron 3 Super is a 120B-parameter open hybrid MoE model, activating just 12B parameters for maximum compute efficiency and accuracy in complex multi-agent applications. Built on a hybrid Mamba-Transformer...

reasoningtool-usereleased 2026-03-11
Legal score · raw
-
not yet benchmarked
Context
1M
max output 16K
Input
$0.08
per 1M tokens
Output
$0.40
per 1M tokens
Suite cost
-
run the suite to see

Benchmark results

TaskCategoryRawJuicedCorrectLatencyCostRan
Hearsay IdentificationEvidence------
Bluebook Citation FormatResearch & Writing------
Federal Civil ProcedureProcedure------
Limitations ArithmeticProcedure------
Contract Clause ClassificationContracts------
Citation Hallucination ResistanceReliability------

Measured by DocketBuster

These numbers come from DocketBuster's own legal battery, not from DocketRouter's suite. Latest run per metric, with n and a 95% Wilson interval where the source reports one. See docketbuster.com/benchmarks.

MetricValuenIntervalMeasured
Statute pinpoint, exact section (no retrieval)13.7% (42/307)30795% CI 10.3% to 18.0%2026-08-24
Statute pinpoint, exact section (with DocketBuster retrieval)75.9% (233/307)30795% CI 70.8% to 80.3%2026-08-24
Say-nothing rate (declines to bluff when the answer is not in the record)100.0%59695% CI 99.4% to 100.0%2026-08-25
Abstained on statute pinpoint18.6% (57/307)307count, no interval reported2026-08-24
Coaching quality (GW-14x, 0 to 8)7.12 / 8-rubric mean, no interval reported2026-08-24

Source files: hard-llm-level-nemotron3-super-120b-statute.json, hard-llm-level-nemotron3-super-120b-statute_rag.json, gw14x-ortier-nemotron3super.json, hard-llm-level-nemotron3-super-120b-saynothing.json. Raw model name in source: nvidia/nemotron-3-super-120b-a12b.