Nemotron 3 Super (120B A12B)
Nemotron 3 Super is a 120B total / 12B active parameter hybrid Mamba-Attention Mixture-of-Experts model optimized for agentic reasoning, coding, planning, tool calling, and long-context analysis. It introduces LatentMoE (projecting tokens into a compressed latent space for expert routing, enabling 4x more experts at the same inference cost), Multi-Token Prediction for native speculative decoding (up to 3x faster generation), and native NVFP4 pretraining on Blackwell. The hybrid architecture interleaves Mamba-2 layers for linear-time sequence processing with strategically placed Transformer attention layers as global anchors, supporting a 1M-token context window. Pre-trained on 25 trillion tokens and post-trained with multi-environment RL across 21 configurations using NeMo Gym/RL with 1.2 million rollouts. Achieves up to 5x higher throughput than previous Nemotron Super and 2.2x higher throughput than GPT-OSS-120B while maintaining comparable accuracy.
Identité
| Organisation | NVIDIA |
|---|---|
| Pays | — |
| Famille | — |
| Type | llm |
| Date de sortie | 2026-03-11 |
| Licence | nvidia open model license agreement |
| Licence commerciale | oui |
| Licence open source | non |
| Poids ouverts | oui |
| Auto-hébergement (DGX Spark) | non — 120 Md > 107 Md |
| Modalités | text |
| Fenêtre de contexte | 1 000 000 |
| KV cache / token (Q8) | — |
| Paramètres totaux | 120 B |
| Paramètres actifs (session) | 12 B |
| Source des paramètres actifs | actifs : nom (heuristique) ; total : nom (heuristique) |
| Curation des paramètres actifs | 2026-11-04 |
| Prix par session | 0.17 $ (1 000 000 tokens entrée + 200 000 sortie, prix le moins cher par fournisseur) |
| Tokens / tâche (Data Players) | 1.2 M |
| Tokens / tâche (Terminal-Bench) | 83.3 M (est.) |
| Ratio Terminal-Bench / Data Players | 69.4× |
| Coût / tâche (Terminal-Bench) | 1.39 $ (est.) |
| Tokens d'entraînement | 25 000 000 000 000 |
| Coupure de connaissance | 2025-06-01 |
| Créé le | 2026-05-07T16:53:33.190523Z |
| Mis à jour le | 2026-11-04T06:04:58.484023Z |
| Source | LLM Stats |
| URL | https://llm-stats.com/models/nemotron-3-super-120b-a12b |
IA Act — preuves fournisseur
| Critère | Note / 5 | Poids |
|---|---|---|
| Documentation technique | 2,5 | 5 % |
| Résumé des données d'entraînement | 0,0 | 35 % |
| Conformité droit d'auteur | 0,0 | 35 % |
| Code of Practice GPAI | 0,0 | 20 % |
| Risque systémique | 0,0 | 5 % |
Curation : 2026-10-23.
IA ActPrix par fournisseur
Statut du modèle (offres + auto-hébergement) : souverain — via Nebul (eurouter).
Toutes sources confondues : catalogue llm-stats, routeurs EUrouter/Cortecs et catalogues directs (OVHcloud, Scaleway, Infomaniak). Les prix sont convertis en USD pour la comparaison ; le prix publié et le taux BCE utilisé sont dans l'infobulle. Les prix inconnus restent indiqués « — ».
| Fournisseur | Source | Souveraineté | Entrée (/M tok) | Sortie (/M tok) | Quantification | Contexte | Latence (s) | Débit (tok/s) | Statut |
|---|---|---|---|---|---|---|---|---|---|
| Nebul | eurouter | souverain | — | — | — | 262 144 | — | — | — |
| DeepInfra | llm-stats | non | 0.09 $ | 0.40 $ | — | — | — | — | active |
Souveraineté qualifiée par offre (modèle × fournisseur) ; survolez le badge pour les critères.
SouverainetéScores par catégorie (top_scores)
| Catégorie | Score |
|---|---|
| chat | 0.60 |
| instruction_following | 0.70 |
| language | 0.80 |
| legal | 0.80 |
| long_context | 0.80 |
| math | 0.70 |
| physics | 0.60 |
| reasoning | 0.60 |
| search | 0.30 |
| finance | 0.80 |
| frontend_development | 0.50 |
| general | 0.60 |
| healthcare | 0.80 |
| agents | 0.30 |
| biology | 0.60 |
| chemistry | 0.60 |
| code | 0.50 |
| communication | 0.60 |
| creativity | 0.70 |
| tool_calling | 0.50 |
| vision | 0.20 |
| writing | 0.70 |
Scores par benchmark
| Benchmark | Catégorie | Score | Max | Normalisé (API) | Vérifié | Auto-déclaré | Daté du |
|---|---|---|---|---|---|---|---|
| aa-lcr | reasoning | 0.58 | 1.00 | 0.58 | non | oui | 2026-11-04T06:04:58.592121Z |
| aime-2025 | math | 0.90 | 1.00 | 0.90 | non | oui | 2026-11-04T06:04:58.501142Z |
| arena-hard-v2 | general | 0.74 | 1.00 | 0.74 | non | oui | 2026-11-04T06:04:58.588804Z |
| bird-sql-(dev) | reasoning | 0.42 | 1.00 | 0.42 | non | oui | 2026-11-04T06:04:58.579383Z |
| browsecomp | reasoning | 0.31 | 1.00 | 0.31 | non | oui | 2026-11-04T06:04:58.573483Z |
| gpqa | reasoning | 0.83 | 1.00 | 0.83 | non | oui | 2026-11-04T06:04:58.520156Z |
| hmmt-2025 | math | 0.95 | 1.00 | 0.95 | non | oui | 2026-11-04T06:04:58.508042Z |
| humanity's-last-exam | reasoning | 0.23 | 1.00 | 0.23 | non | oui | 2026-11-04T06:04:58.534080Z |
| ifbench | instruction_following | 0.73 | 1.00 | 0.73 | non | oui | 2026-11-04T06:04:58.582505Z |
| livecodebench | reasoning | 0.81 | 1.00 | 0.81 | non | oui | 2026-11-04T06:04:58.523303Z |
| mmlu-pro | language | 0.84 | 1.00 | 0.84 | non | oui | 2026-11-04T06:04:58.493515Z |
| mmlu-prox | language | 0.79 | 1.00 | 0.79 | non | oui | 2026-11-04T06:04:58.605846Z |
| multichallenge | chat | 0.55 | 1.00 | 0.55 | non | oui | 2026-11-04T06:04:58.585662Z |
| ruler | long_context | 0.92 | 1.00 | 0.92 | non | oui | 2026-11-04T06:04:58.602618Z |
| scicode | reasoning | 0.42 | 1.00 | 0.42 | non | oui | 2026-11-04T06:04:58.526191Z |
| swe-bench-multilingual | reasoning | 0.46 | 1.00 | 0.46 | non | oui | 2026-11-04T06:04:58.559488Z |
| swe-bench-verified | reasoning | 0.54 | 1.00 | 0.54 | non | oui | 2026-11-04T06:04:58.553062Z |
| tau2-airline | chat | 0.56 | 1.00 | 0.56 | non | oui | 2026-11-04T06:04:58.563674Z |
| tau2-retail | communication | 0.63 | 1.00 | 0.63 | non | oui | 2026-11-04T06:04:58.566600Z |
| tau2-telecom | communication | 0.64 | 1.00 | 0.64 | non | oui | 2026-11-04T06:04:58.570289Z |
| terminal-bench | reasoning | 0.26 | 1.00 | 0.26 | non | oui | 2026-11-04T06:04:58.537110Z |
| terminal-bench-2 | reasoning | 0.31 | 1.00 | 0.31 | non | oui | 2026-11-04T06:04:58.540876Z |
| wmt24++ | language | 0.87 | 1.00 | 0.87 | non | oui | 2026-11-04T06:04:58.608766Z |
Classements
| Catégorie | Rang | Score | Note conservatrice | Benchmarks évalués | Méthode | Daté du | Provenance |
|---|---|---|---|---|---|---|---|
| agents | 190 | — | -0.48 | 3 | trueskill | 2026-10-19 | model_history |
| code | 151 | — | 13.00 | 6 | trueskill | 2026-10-19 | model_history |
Données
Dates de curation (récupération depuis l'API) : models 2026-11-04 · benchmarks 2026-11-04 · scores 2026-11-04 · updates 2026-11-04 · account 2026-11-04 · model_history 2026-11-04 · rankings 2026-11-04