by NVIDIA
Nemotron 3 Nano 30B A3B review, pricing and limits
The entry tier of NVIDIA's Nemotron 3 family, built for cheap, high-throughput agentic AI ahead of the larger Super and Ultra models.
- Best for
- High-volume agentic pipelines
- Not for
- Image, audio, or video understanding
Should you use Nemotron 3 Nano 30B A3B?
Nemotron 3 Nano is the pick for teams building agentic tool-calling pipelines rather than chasing the top reasoning score: giving it tools lifts AIME accuracy from 89.06% to 99.17%, and its MoE architecture keeps inference cheap at scale. It's the wrong choice if you need vision, audio, or frontier-level coding.
Best for
- High-volume agentic pipelines 3.3x the throughput of Qwen3-30B-A3B on identical hardware keeps per-request inference cost low at scale.
- Long-document / long-conversation agents 1M-token context window with 68.2% RULER accuracy at full length beats comparable open models on long-context retention.
- Self-hosted deployments needing commercial rights NVIDIA Open Model License permits commercial use and redistribution of derivatives on a single H100/H200 GPU footprint.
Not for
- Image, audio, or video understanding Text-only model; use the separate Nemotron 3 Nano Omni variant for multimodal input.
- Chasing the top MMLU-Pro score in this size class Qwen3-30B-A3B scores 80.9% vs Nemotron's 78.3%.
- Frontier-level autonomous coding agents 38.76% SWE-Bench (OpenHands) trails dedicated coding-focused frontier models.
Nemotron 3 Nano 30B A3B pricing
- Input, per 1M tokens
- $0.05/1M
- Output, per 1M tokens
- $0.20/1M
- Pricing model
- per-token
- Free tier
- Free tier available via OpenRouter (rate-limited); open weights are also free to self-host.
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.0015 | $0.0002 | $0.0017 |
| Support reply | $0.0001 | $0.0001 | $0.0002 |
| One coding agent run | $0.010 | $0.0040 | $0.014 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Cost examples
- Summarize 100K-token document
- $0.01
- Daily coding agent (1M tokens in / 200K out)
- $0.09
- Customer support chat (2K in / 500 out per turn, 1000 turns)
- $0.20
Input tokens run $0.05 per million and output tokens run $0.20 per million on both DeepInfra and OpenRouter; NVIDIA sets no official per-token price since this is an open-weight model. Self-hosting on an owned or rented NVIDIA H100 or H200 GPU is the lowest per-token cost at high sustained volume.
Self-hosting requirements
- Min vram gb
- 32
- Self-hosted
- Yes
- Recommended GPU
- NVIDIA H100 or H200 (single GPU for BF16/FP8)
What Nemotron 3 Nano 30B A3B can do
Nemotron 3 Nano 30B-A3B is NVIDIA's open-weight hybrid Mamba-Transformer Mixture-of-Experts model built for high-throughput agentic AI. It runs a native 1-million-token context window, far past the 262K default most hosted providers expose, and retains 68.2% accuracy on the RULER long-context benchmark at full length.
What goes in and what comes out
Reads
- text
- tool-calls
Writes
- text
- tool-calls
Capabilities
- Function callingSupported
- Structured outputSupported
- Tool useSupported
- Audio inputNot supported
- Audio outputNot supported
- Code executionNot supported
- Video inputNot supported
- VisionNot supported
- Web browsingNot supported
- Reasoning modes
- reasoning-off, reasoning-on-configurable-budget
- Long context recall
- high
Key features
Hybrid Mamba-Transformer MoE architecture
52 layers interleave Mamba-2 state-space blocks, GQA attention, and MoE routing across 128 experts (6 active per token plus shared experts), the structural reason for its throughput edge over dense models.
Native 1-million-token context window
Extended through continued pretraining at a 512K sequence length; the RULER benchmark shows 68.2% accuracy retained at full context, ahead of Qwen3-30B-A3B at comparable depth.
Configurable reasoning budget
A Reasoning ON/OFF toggle with an adjustable thinking-token budget lets developers trade accuracy for lower inference cost on a per-request basis.
Native tool-calling and structured output
Tool-enabled prompting lifts AIME accuracy from 89.06% to 99.17%, NVIDIA's clearest demonstration that giving the model calculator or code-execution tools closes most of its reasoning gap.
Open commercial license with released safety data
The NVIDIA Open Model License permits commercial redistribution of the model and its derivatives; NVIDIA also published roughly 11,000 labeled agent-safety traces from tool-using workflows.
Known limitations
- Trails Qwen3-30B-A3B on MMLU-Pro (78.3% vs 80.9%).
- SWE-Bench (OpenHands) score of 38.76% is modest versus frontier coding-focused models.
Who it is for
- Teams building high-volume agentic or multi-agent tool-calling pipelines
- Developers who need to self-host an open-weight model with commercial rights
- Teams processing very long documents or conversation histories on a budget
- Cost-sensitive production teams choosing between open-weight 30B-class models
Supported languages
- English
- Spanish
- French
- German
- Japanese
- Chinese
- Korean
- Portuguese
- Italian
- Russian
- Arabic
- Hindi
Nemotron 3 Nano 30B A3B benchmarks
Standout result
Throughput advantage over Qwen3-30B-A3B on identical single-H200 GPU hardware
Source: NVIDIA Nemotron 3 Nano Technical Report, arXiv 2512.20848, December 2025Where it sits
Cheaper than 94% of the 104 GA models with a published price, in the bottom third on SWE-bench Verified (rank 39 of 39), and one of 89 whose vendor states it does not train on customer data.
| Measure | This model | Rank | Median |
|---|---|---|---|
| Blended price (3:1)$ per 1M tokens | $0.088/M | #7 / 104 | peer median $1.73/M |
| SWE-bench Verified% solved | 38.8% | #39 / 39 | peer median 76.5% |
| GPQA Diamond% correct | 73% | #48 / 63 | peer median 87.8% |
Ranks are against GA models on Hokai that publish the same figure; ties share a rank.
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Strengths and limits
Strengths
- Built to run cheaply at scale rather than chase leaderboard position, exactly what high-volume agentic pipelines need.
- Commercially permissive open licensing removes the biggest practical barrier to self-hosting: no royalty or attribution headache for production deployments.
- The tool-enabled reasoning jump is a bigger swing than most models its size manage, and it matters more in practice than the base benchmark score.
- 3.3x higher throughput than Qwen3-30B-A3B on identical single-H200 hardware at 8K in/16K out.
Show 2 more
- 1M-token native context window (RULER-verified 68.2% accuracy at full context), ahead of Qwen3-30B at comparable depth.
- Only ~3.2B of its 31.6B parameters are active per token, cutting real inference cost roughly 10x versus a dense model of similar quality.
Limits
- Text-only: it does not accept image, audio, or video, so multimodal work needs the separate Nemotron 3 Nano Omni variant instead.
- Modest 38.76% SWE-Bench (OpenHands) score versus dedicated coding-focused frontier models.
- Trails Qwen3-30B-A3B on general-knowledge benchmarks like MMLU-Pro, a real tradeoff for the throughput gain.
- Most managed API providers default to a 262K context window rather than the full native ceiling, so the larger window needs explicit configuration.
Nemotron 3 Nano 30B A3B alternatives
The nearest AI models on Hokai for the same job.
| Model | Lab | Context | In / out per 1M | Reads and writes |
|---|---|---|---|---|
| Nemotron 3 Nano 30B A3B This page | NVIDIA | 1M | $0.05 / $0.20 | text → text |
 Gemini 4 HydrogenGoogle DeepMind's November 2026 Gemini 4 model for coding agents, with four thinking levels from low to xhigh and a free tier in AI Studio. | 1M | $1.50 / $7.50 | text, image, audio, video, pdf → text | |
 Qwen4-Max-PreviewAlibaba's November 2026 Qwen 4 flagship preview, open only to Token Plan subscribers, with three reasoning levels from low to xhigh. | Alibaba Cloud | 1M | -- | text, image, video → text |
 Claude Fable 5Superseded by Fable 5.1 and Fable 5.2 but still sold on the Claude API: Anthropic's first generally available Mythos-class model, released June 2026, with a 1M-token context. | Anthropic | 1M | $10 / $50 | text, image, pdf → text |
Nemotron 3 Nano 30B A3B specs
- Context window
- 1.0M tokens
- Architecture
- Hybrid Mamba-2 / Transformer Mixture-of-Experts
- Mixture of experts
- Yes
- Parameters
- 31.6B total, ~3.6B active per token (with embeddings)
- VRAM requirements
- ~63GB (BF16 full precision on a single H200), ~32GB (FP8), lower with 4-bit GGUF quantization
Training data
Pretrained on 25 trillion tokens including 2.5T newly-added Common Crawl tokens, plus 13 million cross-disciplinary post-training samples across 15+ languages, using a Warmup-Stable-Decay learning-rate schedule.
Weight formats
- BF16
- FP8
- GGUF
API endpoints
| Name | Endpoint | Auth |
|---|---|---|
| NVIDIA build.nvidia.com | https://build.nvidia.com/nvidia/nemotron-3-nano-30b-a3b/modelcard | API Key |
| Hugging Face (weights) | https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | None (self-hosted) |
| OpenRouter | https://openrouter.ai/nvidia/nemotron-3-nano-30b-a3b | API Key |
| DeepInfra | https://deepinfra.com/nemotron | API Key |
SDKs
- Python
Prompting patterns
Show 2 prompting patterns
Toggle reasoning mode per request
Set reasoning: on with a thinking-token budget for hard multi-step tasks, off for simple lookups.
Cuts inference cost by up to 60% on tasks that don't need extended chain-of-thought.
Use tool-enabled prompting for math/reasoning tasks
Give the model calculator or code-execution tools for AIME-style problems.
AIME 2025 accuracy jumps from 89.06% (no tools) to 99.17% (with tools).
Known quirks
Hybrid Mamba-Transformer architecture needs updated inference engines
The interleaved Mamba-2 and GQA-Transformer layers are not supported by every older inference server; some engines require recent releases to handle the hybrid architecture correctly.
Workaround: Use vLLM, SGLang, TensorRT-LLM, or llama.cpp at a version released after the model's initial launch, and confirm hybrid-architecture support in the changelog before deploying to production.
Reasoning mode materially changes both cost and accuracy
Turning reasoning on adds visible chain-of-thought tokens that raise both latency and per-request cost, but the accuracy gap versus reasoning-off can be large on harder benchmarks.
Workaround: Set the thinking-token budget explicitly per task type rather than leaving reasoning always-on; route simple tool-calls through reasoning-off and complex multi-step tasks through reasoning-on.
Nemotron 3 Nano 30B A3B facts
- Deployment
- API · Self-hosted · Hugging Face
- Trains on your data
- No
- Licence
- NVIDIA Open Model License · Open weights
- Safety posture
- Configurable
- Released
- 14 Dec 2025
- Current version
- NVIDIA-Nemotron-3-Nano-30B-A3B
- Family
- Nemotron 3
- Variant
- Chat
- Data retention
- Not applicable for self-hosted or open-weight use; retention policy for NVIDIA's build.nvidia.com hosted API was not disclosed in public documentation reviewed.
Content restrictions
- Use must comply with NVIDIA's Trustworthy AI Terms.
- License rights terminate automatically if a user disables or circumvents built-in safety guardrails.
Safety controls
- Rlhf
- Yes
- System prompt
- Yes
- Constitutional AI
- No
- Content filter API
- No
- Moderation endpoint
- No
- Red teaming
- ~11,000 labeled agent-safety traces released publicly; no named third-party red-team partners disclosed
Data governance
- Trains on inputs
- No
- Eu AI act classification
- General-purpose AI model release under NVIDIA's provider obligations
Version history
NVIDIA-Nemotron-3-Nano-30B-A3B · 14 Dec 2025
Initial Nemotron 3 Nano release: 31.6B total / 3.6B active hybrid Mamba-Transformer MoE, 1M context, 78.3% MMLU-Pro.
Nemotron 3 Super (120B) · 11 Mar 2026
Larger Nemotron 3 tier announced at GTC 2026.
Nemotron 3 Ultra (550B) · 04 Jun 2026
Largest Nemotron 3 tier, announced at Computex 2026.
What changed
Questions about Nemotron 3 Nano 30B A3B
How much do you pay for Nemotron 3 Nano 30B A3B?
It runs $0.05 per 1M input tokens and $0.20 per 1M output tokens on both DeepInfra and OpenRouter, among the cheapest rates for a model in this benchmark class, with no official NVIDIA-set price since it's open-weight. A 100K-token document summary costs roughly $0.007 at that rate, and a 1M-in/200K-out coding-agent session runs about $0.09. Self-hosting on a single NVIDIA H100 or H200 GPU removes per-token fees entirely and is the lowest-cost path at sustained high volume.
What do you get on Nemotron 3 Nano 30B A3B's free tier?
NVIDIA doesn't run a hosted free tier of its own, but OpenRouter offers rate-limited free access to the model, and the open weights themselves cost nothing to download and run under NVIDIA's open license. The only real expense in a self-hosted deployment is GPU time, since there's no royalty attached to commercial use.
What are Nemotron 3 Nano 30B A3B's closest competitors?
Qwen3-30B-A3B is the closest open-weight rival in the same 30B-class MoE tier and edges Nemotron out on general-knowledge benchmarks. GPT-OSS-20B is another open competitor NVIDIA benchmarks against, though NVIDIA's own report puts Nemotron's throughput 2.2x ahead of it on identical hardware. Neither rival is currently listed on Hokai; both are named directly in NVIDIA's Nemotron 3 technical report as the comparison set.
How does Nemotron 3 Nano 30B A3B compare to Qwen3-30B-A3B in 2026?
The two trade places depending on what you're optimizing for: Qwen3-30B-A3B scores higher on general knowledge (80.9% vs Nemotron's 78.3% on MMLU-Pro), while Nemotron claims 3.3x its throughput on identical single-H200 hardware at an 8K input / 16K output workload, credited to its hybrid Mamba-Transformer-MoE architecture activating fewer parameters per token. Both figures come from NVIDIA's own technical report rather than independent third-party benchmarking, so treat the throughput claim as a vendor number pending outside verification.
How long does it take to get going with Nemotron 3 Nano 30B A3B?
The fastest way in is an API key from OpenRouter, DeepInfra, or NVIDIA's own build.nvidia.com console, all of which host the model with no local setup required. To self-host instead, download the BF16, FP8, or 4-bit GGUF weights from Hugging Face and serve them with vLLM, SGLang, TensorRT-LLM, or llama.cpp, whichever your existing inference stack already runs. Either path gives you native tool-calling and the model's full context window out of the box, though most managed APIs default to a smaller window until you explicitly request the extended setting.
About Nemotron 3 Nano 30B A3B
Nemotron 3 Nano 30B-A3B is an open-weight language model built by NVIDIA and released on December 14, 2025 as the entry tier of the Nemotron 3 family, ahead of the larger Super (120B, released March 2026) and Ultra (550B, released June 2026) tiers. It was designed to make agentic AI, tool-calling loops carrying long conversation histories, affordable at production scale rather than to chase the largest possible parameter count.
Read the full background
The model interleaves Mamba-2 state-space layers with grouped-query-attention Transformer layers inside a Mixture-of-Experts framework across 52 total layers, with each MoE layer routing to 6 of 128 experts per token plus shared experts active on every token. Of its 31.6 billion total parameters, only about 3.2 billion activate per token, roughly a tenth of the full weight count, the architectural lever behind its throughput. On other benchmarks from NVIDIA's technical report (arXiv 2512.20848), it scores 73.04% on GPQA, 68.25% on LiveCodeBench v6, and 71.51% on IFBench prompt-following.
Training used a 25-trillion-token pretraining corpus, including 2.5 trillion newly-added Common Crawl tokens, plus 13 million post-training samples across 15+ languages, with a Warmup-Stable-Decay learning-rate schedule; NVIDIA has not disclosed an exact training-data cutoff date. The model ships under the NVIDIA Open Model License, a commercially permissive custom license whose rights terminate automatically if a user disables the model's built-in safety guardrails.
It's available through NVIDIA's own build.nvidia.com hosted API, Hugging Face weight downloads, OpenRouter, and DeepInfra, or self-hosted via vLLM, SGLang, TensorRT-LLM, llama.cpp, LM Studio, or Unsloth. NVIDIA has signaled a coalition-built Nemotron 4 family as the next generation, with the Super and Ultra tiers already extending the same architecture through mid-2026.
Sources and guides
Official sources
- Nemotron 3 Nano 30B A3B model page Vendor
- Licence text Licence
Specifications on this page (context window, token prices, benchmark scores, licence) are read from NVIDIA's own model card and documentation. The verdict is Hokai's reading of them, not the provider's claim.
Lists that include Nemotron 3 Nano 30B A3B
The provider
More AI Models on Hokai
- Claude Fable 5.1Superseded by Claude Fable 5.2 in November 2026 but still sold on the Claude API: Anthropic's second Mythos-class model, released Sept 1, 2026, for long-horizon coding and research.
Gemini Decision 1Google DeepMind's decision model scores many questions over one multimodal input of up to 262,144 tokens and bills zero output tokens.
Shieldstral 1.0Mistral's open-weight multimodal safety classifier, released 4 August 2026, that screens text and images against policies you write.
- Claude Mythos 5.2Anthropic's November 2026 Mythos model: Fable 5.2's weights with relaxed cyber and biology safeguards for verified teams, with a 1M-token context.
ERNIE-5.5-Flash-PreviewBaidu's text-only preview mixture-of-experts model, served through Qianfan on OpenRouter with a 1M-token context window and tool calling.
- Gemini 4 CarbonGoogle DeepMind's Gemini 4 Carbon, open to paying Gemini API developers since 10 November: a reasoning model for agentic coding, research and cyber defense.