Nemotron 3 Super
nvidia/nemotron-3-super-120b-a12bNVIDIA 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.
Standard pricing
Run this model at its standard rate. Cheap mode currently has no separate discount.
API routing ceilings, including your 5% Free-plan fee. Included usage applies first, then prepaid credits. Cached tokens and tools can have separate charges.
Other models with discounted rates
See all discounts →These are different models. Cheap mode may be slower; discounts compare each model’s own standard rate.
Choose how to use Nemotron 3 Super
Playground
Try prompts in your browser, create an agent, and see the cost of each reply. Your selected model carries into setup.
Model capabilities below describe the underlying model. Available playground features depend on its provider.
API
Connect your app with a Gizmo API key. Choose fastest or cheapest, stream text responses, and inspect timing, tokens, and charges in request logs.
A current API route is listed for this exact model. Gateway configuration is required for live calls.
Provider price + a transparent platform fee
| Rate | Provider baseline | Free plan · 5% fee | Paid plan · 4% fee |
|---|---|---|---|
| Input / 1M tokens | $0.075 | $0.07875 | $0.078 |
| Output / 1M tokens | $0.4 | $0.42 | $0.416 |
| API input / 1M tokens | $0.075 | $0.07875 | $0.078 |
| API output / 1M tokens | $0.4 | $0.42 | $0.416 |
Legacy chat image rates use the catalog baseline. The dedicated image playground shows separate estimates and settles the provider’s metered usage. API rates are current routing ceilings; final API charges use provider-reported cost plus your fee. Cheapest routing compares eligible OpenRouter and Cheaper Inference rates for the same model; savings are not guaranteed. Cached input or other provider charges may differ. Compare all model rates and verified discounts →
How your plan pays for usage
API and playground share one included allowance. Usage including the fee draws from that allowance first, then your prepaid credits. Plans bill monthly; included usage resets weekly: Free $1, Starter $5, Growth $25, Scale $50.
A $1 provider charge becomes $1.05 on Free or $1.04 on a paid plan before included usage is applied. Plan eligibility still applies in the playground.
Free models and limits
Zero-cost models stay $0, including the fee, and do not consume your dollar allowance. Free accounts can make 1,000 free-model API requests per calendar month, resetting in UTC. Admitted requests count even if the provider fails.
Free includes 20 API requests/minute and 6 playground messages/minute. Paid plans raise these limits. See plan limits.
Before a text API request, Gizmo temporarily reserves included usage and, if necessary, prepaid credit for the model’s full context plus your maximum output. Image jobs reserve twice their displayed estimate. Unused funds return after metering. This hold can exceed the final charge; unconfirmed usage remains marked pending.
Overview
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 Mixture-of-Experts architecture with multi-token prediction (MTP), it delivers over 50% higher token generation compared to leading open models.
The model features a 1M token context window for long-term agent coherence, cross-document reasoning, and multi-step task planning. Latent MoE enables calling 4 experts for the inference cost of only one, improving intelligence and generalization. Multi-environment RL training across 10+ environments delivers leading accuracy on benchmarks including AIME 2025, TerminalBench, and SWE-Bench Verified.
Fully open with weights, datasets, and recipes under the NVIDIA Open License, Nemotron 3 Super allows easy customization and secure deployment anywhere — from workstation to cloud.
Benchmarks
Artificial Analysis indices independent, 0–100
Benchmarks from the model card
| Benchmark | Nemotron-3-Super | Nemotron-3-Super FP8 | Nemotron-3-Super NVFP4 |
|---|---|---|---|
| General Knowledge | |||
| MMLU-Pro | 83.73 | 83.63 | 83.33 |
| Reasoning | |||
| HMMT Feb25 (with tools) | 94.73 | 94.38 | 95.36 |
| GPQA (no tools) | 79.23 | 79.36 | 79.42 |
| LiveCodeBench (v6 2024-08↔2025-05) | 78.69 | 78.44 | 78.44 |
| LiveCodeBench (v5 2024-07↔2024-12) | 81.19 | 80.99 | 80.56 |
| SciCode (subtask) | 42.05 | 41.38 | 40.83 |
| HLE (no tools) | 18.26 | 17.42 | 17.42 |
| Agentic | |||
| Terminal Bench (hard subset) | 25.78 | 26.04 | 24.48 |
| TauBench V2 | |||
| Airline | 56.25 | 56.25 | 54.75 |
| Retail | 62.83 | 63.05 | 63.38 |
| Telecom | 64.36 | 63.93 | 63.27 |
| Average | 61.15 | 61.07 | 60.46 |
| Chat & Instruction Following | |||
| IFBench (prompt) | 72.58 | 72.32 | 73.30 |
| Scale AI Multi-Challenge | 55.23 | 54.35 | 52.8 |
| Arena-Hard-V2 (Hard Prompt) | 73.88 | 76.06 | 76.00 |
| Long Context | |||
| AA-LCR | 58.31 | 57.69 | 58.06 |
| RULER-500 @ 128k (500 samples per task) | 96.79 | 96.85 | 95.99 |
| RULER-500 @ 256k (500 samples per task) | 96.60 | 96.33 | 96.52 |
Base Pre-Training Corpus (Nemotron 3 Foundation) from the model card
| Dataset Collection | Token Counts | Description |
|---|---|---|
| Nemotron-CC-v2 & v2.1 | 9.13T | A massive collection of English web data filtered from Common Crawl, including 2.5T+ tokens of new organic, translated, and synthetically rephrased content. |
| Nemotron-CC-Code-v1 | 427.9B | High-quality code tokens extracted from Common Crawl using the Lynx + LLM pipeline to preserve structure and equations. |
| Nemotron-Pretraining-Code-v1 & v2 | 1.09T | Curated GitHub code references with multi-stage filtering, deduplication, and large-scale synthetic code data. |
| Nemotron-CC-Math-v1 | 133.3B | High-quality math pre-training dataset preserving LaTeX formatting and mathematical structures. |
| Nemotron-Pretraining-Specialized-v1 | 336.4B | Synthetic datasets targeting specialized domains such as STEM reasoning and scientific coding. |
Public Datasets from the model card
| Dataset | Collection Period |
|---|---|
| GSM8K | 4/23/2025 |
| CC-NEWS | 4/23/2025 |
| Common Crawl | 4/23/2025 |
| Wikimedia | 4/23/2025 |
| Bespoke-Stratos-17k | 4/23/2025 |
| tigerbot-kaggle-leetcodesolutions-en-2k | 4/23/2025 |
| glaive-function-calling-v2 | 4/23/2025 |
| APIGen Function-Calling | 4/23/2025 |
| LMSYS-Chat-1M | 4/23/2025 |
| Open Textbook Library \- CC BY-SA & GNU subset and OpenStax \- CC BY-SA subset | 4/23/2025 |
| Advanced Reasoning Benchmark, tigerbot-kaggle-leetcodesolutions-en-2k, PRM800K, and SciBench | 4/23/2025 |
| FineWeb-2 | 4/23/2025 |
| Court Listener | Legacy Download |
| peS2o | Legacy Download |
| OpenWebMath | Legacy Download |
| BioRxiv | Legacy Download |
| PMC Open Access Subset | Legacy Download |
| OpenWebText2 | Legacy Download |
| Stack Exchange Data Dump | Legacy Download |
| PubMed Abstracts | Legacy Download |
| NIH ExPorter | Legacy Download |
| arXiv | Legacy Download |
| BigScience Workshop Datasets | Legacy Download |
| Reddit Dataset | Legacy Download |
Private Non-publicly Accessible Datasets of Third Parties from the model card
| Dataset | Model(s) used |
|---|---|
| Global Regulation | Unknown |
| TAUS Translation Memory | Unknown |
| Scale HLE | Unknown |
| HackerRank Coding | Unknown |
| RL data for Search | Gemini 3; GPT-5 * |
Capabilities
Reasoning is on by default with effort levels medium · low (default medium).
Sampling controls: temperature, top_p, top_k, min_p, max_tokens, frequency_penalty, presence_penalty, repetition_penalty. Defaults: top_k null, top_p 0.95, temperature 1, presence_penalty null, frequency_penalty null, repetition_penalty null.
Pricing on Gizmo
These are provider baseline rates before Gizmo’s 5% Free-plan or 4% paid-plan fee. Your included usage covers charges first; overage uses prepaid credits. See the fee-inclusive comparison above.
Provider baseline rates
- Input
- $0.075 / 1M
- Output
- $0.4 / 1M
Examples before the platform fee
- Typical chat turn
- $0.00035 2k in · 500 out
- $1 of input
- ≈ 13.33M tokens
- $1 of output
- ≈ 2.5M tokens
- Full context, once
- $0.0786
Providers and pricing
One way to run Nemotron 3 Super on Gizmo.
| Provider | Route | Input / 1M | Output / 1M | Context | Uptime 24h |
|---|---|---|---|---|---|
| DeepInfra · bf16Lowest price | OpenRouter | $0.075 | $0.40 | 262k | 100% |
Rates are each provider’s, per million tokens, before Gizmo’s 5% (Free) or 4% (paid) fee. “Cheapest” routing starts at the top of this list; “Fastest” starts with the model’s maker.
More from NVIDIA
Nemotron 3.5 Super VLNVIDIA · 1.05M ctx$0.15 · $0.5Nemotron 3.5 LightningNVIDIA · 262k ctx$0.07 · $0.2Nemotron 3.5 Lightning (free)NVIDIA · 1M ctxFreeNemotron 3 UltraNVIDIA · 262k ctx$0.6 · $2.40Nemotron 3 Ultra (free)NVIDIA · 1M ctxFreeNemotron 3 Nano 30B A3BNVIDIA · 262k ctx$0.06 · $0.24Similar price, other creators
Run Nemotron 3 Super in under a minute
Create your account
Sign up with just an email. Start on the free tier with no card, or pick a paid plan for a bigger weekly allowance and $25 in bonus credits.
Your agent is ready
Nemotron 3 Super is preselected as the brain of your first agent. Give it a name and a system prompt, or keep the defaults.
Chat, research, create
Live web research and image generation are built in. Every reply shows what it cost, and you can switch models anytime.