# Nemotron Nano 2 > Source: https://aiwiki.ai/wiki/nemotron_nano_2 > Updated: 2026-10-28 > Categories: AI Models, Large Language Models, Open Source AI > License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)" > Cite as: AI Wiki. "Nemotron Nano 2." aiwiki.ai, 28 Oct 2026. https://aiwiki.ai/wiki/nemotron_nano_2 > From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution. **Nemotron Nano 2** is a family of small, open-weight [reasoning](https://aiwiki.ai/wiki/reasoning_models) [language model](https://aiwiki.ai/wiki/language_model)s released by [NVIDIA](https://aiwiki.ai/wiki/nvidia) on August 18, 2025, built on a hybrid [Mamba-2](https://aiwiki.ai/wiki/mamba_2) [state space model](https://aiwiki.ai/wiki/state_space_model) plus Transformer architecture that replaces most [self-attention](https://aiwiki.ai/wiki/self_attention) layers with Mamba-2 layers to generate long reasoning traces faster. Its flagship checkpoint, NVIDIA-[Nemotron](https://aiwiki.ai/wiki/nemotron)-Nano-9B-v2, runs a 128,000-token [context window](https://aiwiki.ai/wiki/context_window) on a single mid-range [data center](https://aiwiki.ai/wiki/data_center) GPU and, according to NVIDIA, delivers up to roughly 6 times higher inference throughput than the similarly sized Qwen3-8B while matching or exceeding its reasoning accuracy.[1][2] NVIDIA released the model weights together with most of the pre-training corpus, the Nemotron-Pre-Training-Dataset-v1, making Nemotron Nano 2 one of the more openly documented small reasoning model releases of 2025.[2][3]
Nvidia logo
Nvidia logo. Source: Wikimedia Commons, Apache License 2.0.
"Nano" denotes the small and efficient tier of the Nemotron program, intended to run a full 128,000-token context window on a single mid-range data center GPU. The flagship aligned model, NVIDIA-Nemotron-Nano-9B-v2, has about 9 billion parameters and was compressed and distilled from a 12-billion-parameter base model, NVIDIA-Nemotron-Nano-12B-v2-Base.[1][4] ## What is Nemotron Nano 2? Nemotron Nano 2 was announced on August 18, 2025, with a companion technical report, "NVIDIA Nemotron Nano 2: An Accurate and Efficient Hybrid Mamba-Transformer Reasoning Model," posted to [arXiv](https://aiwiki.ai/wiki/arxiv) (2508.14444) and credited to a team of 215 NVIDIA authors.[1] The release comprised three checkpoints on [Hugging Face](https://aiwiki.ai/wiki/hugging_face): the aligned reasoning model NVIDIA-Nemotron-Nano-9B-v2, plus two base checkpoints, NVIDIA-Nemotron-Nano-9B-v2-Base and NVIDIA-Nemotron-Nano-12B-v2-Base.[2][4] The family targets a specific deployment regime: agentic and reasoning tasks where a model must generate long chains of intermediate tokens before producing a final answer. Because the cost of generating those tokens dominates inference, NVIDIA prioritized decode-time throughput, using the Mamba-heavy hybrid backbone to lower the per-token compute and memory cost relative to attention.[1][2] The 9B model supports a maximum context length of 128,000 tokens and, after compression, can serve that full context on a single NVIDIA A10G GPU (22 GiB of memory) in [bfloat16](https://aiwiki.ai/wiki/bfloat16) precision.[2][4] The models support toggleable reasoning. By default the aligned model emits an explicit reasoning trace, which a developer can disable with a control token, and it exposes a configurable "thinking budget" that caps how many internal reasoning tokens are spent before the model must answer.[4][5] Supported languages include English, German, Spanish, French, Italian, Japanese, Korean, Portuguese, Russian, and Chinese, among others.[4] ## Who developed Nemotron Nano 2? Nemotron is NVIDIA's family of open language models and associated datasets, spanning model lines such as [Nemotron 3](https://aiwiki.ai/wiki/nemotron_3), [Llama Nemotron](https://aiwiki.ai/wiki/llama_nemotron) (built on Meta's Llama base models), and the efficiency-focused research model [Jet-Nemotron](https://aiwiki.ai/wiki/jet_nemotron). The Nano 2 architecture is derived from [Nemotron-H](https://aiwiki.ai/wiki/nemotron_h), an earlier NVIDIA hybrid Mamba-Transformer line that established the approach of substituting Mamba-2 layers for most attention layers.[1][2] NVIDIA has positioned the Nemotron program around openness, releasing not only weights but also large training datasets and recipes. The Nano 2 release continued this by publishing the Nemotron-Pre-Training-Dataset-v1, a corpus that NVIDIA describes as a major portion of the data used to train the models, alongside post-training data.[2][3] The dataset builds on NVIDIA's earlier [Nemotron-CC](https://aiwiki.ai/wiki/nemotron_cc) web-crawl curation work. ## What is the hybrid Mamba-Transformer architecture? The Nano 2 models are built on the Nemotron-H hybrid design. In a conventional Transformer, every block uses self-attention; in the Nano 2 backbone, the great majority of those blocks are replaced with Mamba-2 state space model layers, which scale linearly with sequence length and maintain a fixed-size recurrent state rather than a growing key-value cache.[1][2] A small number of attention layers are kept to preserve the in-context retrieval behavior that pure state space models struggle with. The 12-billion-parameter base model, Nemotron-Nano-12B-v2-Base, is reported to use 62 layers in total: 6 self-attention layers, 28 feed-forward (MLP) layers, and 28 Mamba-2 layers, with a hidden dimension of 5,120, a feed-forward intermediate dimension of 20,480, and [grouped-query attention](https://aiwiki.ai/wiki/grouped_query_attention) using 40 query heads and 8 key-value heads.[6] The base model was pre-trained on roughly 20 trillion tokens using an FP8 training recipe.[1][4] To produce the deployable 9B model, NVIDIA applied the [Minitron](https://aiwiki.ai/wiki/minitron) compression strategy, which combines structured [pruning](https://aiwiki.ai/wiki/pruning) with [knowledge distillation](https://aiwiki.ai/wiki/knowledge_distillation) from the larger model. The resulting NVIDIA-Nemotron-Nano-9B-v2 keeps 56 of the original layers, including just 4 attention layers, with the embedding dimension pruned from 5,120 to 4,480 and the feed-forward intermediate dimension pruned from 20,480 to 15,680.[6] This compression is what allows the model to fit a 128K-token context on a single A10G GPU.[1][2] ### Specifications | Attribute | Detail | |---|---| | Developer | [NVIDIA](https://aiwiki.ai/wiki/nvidia) | | Announced | August 18, 2025[1] | | Architecture | Hybrid Mamba-2 + Transformer attention ([Nemotron-H](https://aiwiki.ai/wiki/nemotron_h) lineage)[1][2] | | Aligned model | NVIDIA-Nemotron-Nano-9B-v2 (~9B parameters)[4] | | Base models | Nemotron-Nano-12B-v2-Base, Nemotron-Nano-9B-v2-Base[2] | | Base pre-training | ~20 trillion tokens, FP8 precision[1][4] | | 12B base layers | 62 total: 6 attention, 28 MLP, 28 Mamba-2; hidden 5,120; GQA 40 query / 8 KV heads[6] | | 9B model layers | 56 total, including 4 attention; hidden 4,480[6] | | Compression method | [Minitron](https://aiwiki.ai/wiki/minitron) pruning and [distillation](https://aiwiki.ai/wiki/knowledge_distillation)[1] | | Context length | 128,000 tokens[2][4] | | Single-GPU target | 128K context on one NVIDIA A10G (22 GiB), bfloat16[2][4] | | Reasoning control | Toggleable thinking, configurable thinking-token budget[4][5] | | Throughput claim | Up to ~6x vs Qwen3-8B in reasoning settings (NVIDIA)[1][2] | | Weights license | NVIDIA Open Model License Agreement[4] | | Dataset license | Creative Commons Attribution 4.0 (CC BY 4.0)[3] | ## Is Nemotron Nano 2 open source, and what data was released? The headline checkpoint, NVIDIA-Nemotron-Nano-9B-v2, is an aligned, post-trained reasoning model that NVIDIA states is "ready for commercial use" under the NVIDIA Open Model License Agreement.[4] The two base checkpoints are provided for further fine-tuning: Nemotron-Nano-12B-v2-Base is the full [pre-trained model](https://aiwiki.ai/wiki/pre-trained_model) before compression and alignment, and Nemotron-Nano-9B-v2-Base is the pruned base prior to alignment.[2][4] A central feature of the release is the openness of the training data. NVIDIA published the Nemotron-Pre-Training-Dataset-v1, which it describes as comprising about 6.6 trillion tokens of web crawl, mathematics, code, [supervised fine-tuning](https://aiwiki.ai/wiki/supervised_fine-tuning), and multilingual question-answer data.[2][3] The collection includes several named components: Nemotron-CC-v2 (a multilingual web-crawl dataset with synthetic question-answer rephrasings), Nemotron-CC-Math-v1 (about 133 billion tokens of mathematics-focused data), Nemotron-Pretraining-Code-v1 (a curated code corpus), and Nemotron-Pretraining-SFT-v1 (synthetic instruction-tuned data).[2][3] The dataset is distributed under the Creative Commons Attribution 4.0 (CC BY 4.0) license, while the model weights are governed by the separate NVIDIA Open Model License.[3][4] NVIDIA reported data-quality gains from these components, for example improvements on the MATH benchmark of roughly +4.8 to +12.6 points over strong baselines from its math data, +4.6 to +14.3 points on MBPP+ for code generation, and about +10.0 on Global-MMLU from its multilingual question-answer data versus using multilingual [Common Crawl](https://aiwiki.ai/wiki/common_crawl) alone.[2] These figures are NVIDIA's own ablation results. ## How fast is Nemotron Nano 2, and how accurate is it? NVIDIA presents Nemotron Nano 2 as achieving leading accuracy among open models of comparable size while running substantially faster. In the technical report, the 9B model is benchmarked primarily against Qwen3-8B, an open small model from Alibaba's [Qwen3](https://aiwiki.ai/wiki/qwen_3) family that NVIDIA identifies as the strongest comparably sized baseline.[1][2] In its own announcement, NVIDIA states that Nemotron-Nano-9B-v2 "achieves comparable or better accuracies on complex reasoning benchmarks than the leading comparably sized open model Qwen3-8B at up to 6x higher throughput."[2] With reasoning enabled, NVIDIA's published scores for NVIDIA-Nemotron-Nano-9B-v2 include MATH500 at 97.8 percent, [AIME 2025](https://aiwiki.ai/wiki/aime_2025) at 72.1 percent, GPQA at 64.0 percent, [LiveCodeBench](https://aiwiki.ai/wiki/livecodebench) at 71.1 percent, BFCL v3 (tool use) at 66.9 percent, IFEval (instruction strict) at 90.3 percent, and RULER at 128K context at 78.9 percent.[4] In head-to-head comparisons reported by NVIDIA, the model scored 56.67 versus 20.00 for Qwen3-8B on [AIME 2024](https://aiwiki.ai/wiki/aime_2024) (pass@32), 64.48 versus 59.61 on GPQA-Diamond, and 97.75 versus 96.3 on MATH-500.[6] NVIDIA characterizes the overall result as on-par or better accuracy than Qwen3-8B while delivering up to 6 times higher inference throughput in reasoning settings such as 8,000 input and 16,000 output tokens, with reported speedups varying roughly from 3x to 6x depending on the workload.[1][6] All of these are vendor-reported numbers and have not been independently audited here. ## How does reasoning control work in Nemotron Nano 2? A distinguishing capability is fine-grained control over reasoning. The model defaults to producing an explicit reasoning trace, activated by a `/think` control token, and a `/no_think` token disables the trace for latency-sensitive or simple requests.[4][5] Developers can additionally set a thinking budget through a `max_thinking_tokens` parameter, capping the internal reasoning length before the model must commit to an answer; per the [model card](https://aiwiki.ai/wiki/model_card), "the thinking budget allows developers to keep accuracy high and meet response-time targets."[4][5] NVIDIA reports that this behavior was instilled during post-training by including a fraction of the data, about 5 percent, with deliberately truncated reasoning traces, teaching the model to produce a useful answer even when its thinking is cut short.[2][5] ## Why does Nemotron Nano 2 matter? Nemotron Nano 2 is notable on two fronts. Architecturally, it is one of the more prominent demonstrations that a hybrid [Mamba](https://aiwiki.ai/wiki/mamba)-Transformer model can match strong Transformer reasoning models at the small-model scale while offering a large throughput advantage, an advantage that grows with the length of the reasoning traces these models generate. By keeping only a few attention layers, the design avoids the quadratic attention cost and the large key-value cache that constrain long-context decoding in conventional small models.[1][2] On openness, the release goes further than most open-weight small models by publishing the majority of the pre-training corpus under a permissive license, in addition to the weights. This places Nemotron Nano 2 among the more reproducible small reasoning model efforts of 2025, comparable in spirit to other open small-model families such as Qwen3, [Microsoft](https://aiwiki.ai/wiki/microsoft)'s [Phi](https://aiwiki.ai/wiki/phi_4), and [Google](https://aiwiki.ai/wiki/google)'s [Gemma](https://aiwiki.ai/wiki/gemma), while distinguishing itself through the combination of a state-space hybrid backbone and open data.[2][3] The work also extended NVIDIA's Nemotron-H and Minitron research, and the hybrid Mamba-Transformer approach was carried forward into subsequent Nemotron releases.[1][6] ## References 1. NVIDIA. "NVIDIA Nemotron Nano 2: An Accurate and Efficient Hybrid Mamba-Transformer Reasoning Model." arXiv:2508.14444, August 2025. https://arxiv.org/abs/2508.14444 2. NVIDIA ADLR. "NVIDIA Nemotron Nano 2 and the Nemotron Pretraining Dataset v1." NVIDIA Research, August 18, 2025. https://research.nvidia.com/labs/adlr/NVIDIA-Nemotron-Nano-2/ 3. NVIDIA. "Nemotron-Pre-Training-Dataset-v1." Hugging Face dataset collection, 2025. https://huggingface.co/collections/nvidia/nemotron-pre-training-dataset 4. NVIDIA. "NVIDIA-Nemotron-Nano-9B-v2 (model card)." Hugging Face, 2025. https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2 5. NVIDIA. "Supercharge Edge AI With High-Accuracy Reasoning Using NVIDIA Nemotron Nano 2 9B." Hugging Face Blog, 2025. https://huggingface.co/blog/nvidia/supercharge-ai-reasoning-with-nemotron-nano-2 6. NVIDIA. "NVIDIA Nemotron Nano 2: An Accurate and Efficient Hybrid Mamba-Transformer Reasoning Model" (HTML technical report). arXiv, 2025. https://arxiv.org/html/2508.14444v2