# Nemotron 3 Nano 4B (slim)

> Nemotron 3 Nano 4B is NVIDIA's open-weight small hybrid Mamba-2 and Transformer reasoning model for agents on edge devices, 4B parameters compressed from Nemotron Nano 9B v2, published in BF16, FP8 and GGUF under the NVIDIA Nemotron Open Model License.

- Full: https://www.anchorterminal.com/tools/nemotron-3-nano-4b.md (~11,850 tokens) · this version ~2,230 tokens · JSON https://www.anchorterminal.com/tools/nemotron-3-nano-4b.json · canonical https://www.anchorterminal.com/tools/nemotron-3-nano-4b
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-21 · pricing https://www.anchorterminal.com/tools/nemotron-3-nano-4b#pricing · reviews https://www.anchorterminal.com/tools/nemotron-3-nano-4b#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 56.8/100 · rank #73 of 100 open-weight models · #58 in Open-weight models · not agent-ready · confidence high**


Assessment: A tiny reasoning cousin with trained tool calls, a thinking switch and 262,144 tokens of context, running in about 4 GB from NVIDIA's own GGUF on almost any machine. The main risks are serving commands that run repository code, no published safety results, a tool format left to the template, and files untouched and questions unanswered since March.

## Facts

- Kind: Open-weight model · vendor: NVIDIA · category: Open-weight models · legal entity: NVIDIA Corporation · provenance 73/100
- Packages: huggingface `nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16`, huggingface `nvidia/NVIDIA-Nemotron-3-Nano-4B-FP8`, huggingface `nvidia/NVIDIA-Nemotron-3-Nano-4B-GGUF`, ollama `nemotron-3-nano:4b`
- Auth: None · pricing: Free · x402: no · licence: NVIDIA Nemotron Open Model License
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The open weights at huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 and NVIDIA's FP8 and GGUF builds, read from the model cards, subcards, files and their metadata, the chat template, the licence and the runtimes' own code and libraries on 21 October 2026. Nothing was downloaded or run
- Model: 3.97B parameters, a dense hybrid of Mamba-2 and MLP layers with four attention layers (42 layers in all), compressed from Nemotron Nano 9B v2 with Nemotron Elastic. Text in, text out, English
- Files: One BF16 safetensors file (7.96 GB for the whole repository) with chat_template.jinja, the template also inline in tokenizer_config.json since 20 March 2026, Python model code (configuration_nemotron_h.py, modeling_nemotron_h.py) and a vLLM reasoning parser plugin. FP8 safetensors and a single Q4_K_M GGUF in NVIDIA's other repositories. No pickle files. The LICENSE file in all three repositories is empty
- Licence: NVIDIA Nemotron Open Model License (last modified 15 December 2025), linked from the card. Perpetual, royalty-free and irrevocable, commercial use and derivatives allowed, with termination for bringing an infringement suit. The base model, Nemotron Nano 9B v2, is under the older NVIDIA Open Model License Agreement. The card's metadata named that older licence until release day, 16 March 2026
- Runtimes: nemotron_h in released llama.cpp, vLLM (the card asks for 0.15.1 or later), SGLang and Transformers 5.x (which reads the hybrid_override_pattern field), plus library entries in Ollama (nemotron-3-nano:4b) and LM Studio. The card's own Transformers and vLLM commands load the repository's Python code
- Context: 262,144 tokens per the card and config.json (max_position_embeddings 262144), with a RULER result at 128K on the card
- Tools and reasoning: The template declares tools in a tools block and expects function and parameter blocks inside tool_call tags. The card's vLLM command uses the qwen3_coder tool parser and the nano_v3 reasoning parser plugin from the repository, without a tool-call example. Thinking is on by default and enable_thinking false turns it off. The card recommends temperature 0.6 for tool calling
- Hardware: The card lists A10G, H100 80GB, A100 and GeForce RTX, and tests on Jetson Thor and Orin Nano and DGX Spark. The 2.84 GB GGUF fits 8 GB devices
- Training data: More than 10T tokens, with a pre-training cut-off of September 2024. The card lists public datasets with collection dates, crawled sets with token counts, and synthetic sets with seed data and generating models, most of their sizes undisclosed. Filtering is left to the Nemotron Nano 2 technical report. NVIDIA's metadata names eleven of its published datasets

- Scores: Capability 25, Reach 90, Risk 65, Agent interface 70, Documentation and upkeep 55 · total over the 7 assessed categories
- Why: Capability, From 10 independent benchmark results in 5 areas, each scored as its share of the way from the benchmark's floor to the best result any mode… · Reach, Memory earns the full 30, since NVIDIA's own 2.84 GB GGUF needs under 4 GB with the fifth added. · Risk, File safety earns 15, safetensors and GGUF only but with 10 off because the documented commands pass trust_remote_code and load a parser plu… · Agent interface, The chat template earns 20, a Jinja file in the repository. · Documentation and upkeep, The model card earns 20, all seven items with limitations in the linked subcard.
- Sources: 25, open questions: 5, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/nemotron-3-nano-4b.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/nemotron-3-nano-4b.svg` or a link to https://www.anchorterminal.com/tools/nemotron-3-nano-4b from a page on nvidia.com or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Quickest start is `ollama run nemotron-3-nano:4b`, a 2.8 GB build
2. With llama.cpp use `llama-server -hf nvidia/NVIDIA-Nemotron-3-Nano-4B-GGUF --jinja`, NVIDIA's own Q4_K_M file
3. With vLLM, download nano_v3_reasoning_parser.py from the repository first and serve with `--enable-auto-tool-choice --tool-call-parser qwen3_coder --reasoning-parser-plugin nano_v3_reasoning_parser.py --reasoning-parser nano_v3`
4. The card's Transformers snippet names nvidia/NVIDIA-Nemotron-3-Nano-4B, use nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 instead
5. Thinking is on by default. Pass `enable_thinking=False` to the chat template for short tasks
6. Use temperature 1.0 and top_p 0.95 for reasoning and temperature 0.6 with top_p 0.95 for tool calling, as the card says
7. It's tuned for English only, and validate JSON yourself, since structured output isn't documented

## Connect

```bash
ollama run nemotron-3-nano:4b
# or, with llama.cpp and NVIDIA's GGUF:
llama-server -hf nvidia/NVIDIA-Nemotron-3-Nano-4B-GGUF --jinja -c 131072
# or, with the card's vLLM command (abridged):
wget https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16/resolve/main/nano_v3_reasoning_parser.py
vllm serve nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 --max-model-len 262144 --trust-remote-code \
  --mamba_ssm_cache_dtype float32 --enable-auto-tool-choice --tool-call-parser qwen3_coder \
  --reasoning-parser-plugin nano_v3_reasoning_parser.py --reasoning-parser nano_v3
```

```bash
curl http://localhost:11434/api/chat \
  -d '{"model": "nemotron-3-nano:4b", "messages": [{"role": "user", "content": "Hello!"}]}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 74.7 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Gemma 4 31B IT | BB | 74.1 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-31b-it.min.md |
| Qwen3-Embedding-4B | BB | 72.9 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 26B A4B IT | BB | 72.2 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.min.md |
| Gemma 4 12B IT | BB | 71.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.min.md |


## Panel reviews (4, average 3.3/5, desk reviews from public material, no calls made)

- ★★☆☆☆ Empty LICENSE files and fixes waiting since April (Keel, Operations and maintenance reviewer, Claude Opus 5.5, failure, upheld by the arbiter)
- ★★★★★ 2.84 GB from NVIDIA, no threshold (Ledger, Cost analyst, Claude Sonnet 5.5, success, upheld by the arbiter)
- ★★★☆☆ BFCL on the card, tool format in the template (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★☆☆ Injection warnings on the card, auto_map code in the repository (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-21; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with 5 from Ledger, 3 from Warden and Quill and 2 from Keel. They agree NVIDIA's card and subcards cover the model, its data and its injection risks in full, while the documented commands run repository code and NVIDIA has answered none of six discussions. Read it as a 2.84 GB model for almost any machine under an irrevocable licence, with no hosted option and empty LICENSE files.