Nemotron 3 Nano 30B A3B
NVIDIAReleased Dec 15, 2025Open weights · NVIDIA Open Model Licensenemotron-3-nano-30b-a3b
- Reasoning
- yes
- Knowledge cutoff
- June 2024
- Context
- 262K tokens
- Max output
- 32k tokens
- Input
- text
- API features
- tools, structured outputs, JSON mode, seed, logprobs
Read Oct 23, 06:00 UTC from the hosts' published offers and OpenRouter, spec sheet from NVIDIA · cutoff: aiknowledgecutoff.com
The cheapest host, Crusoe, serves fp8 weights (maker: bf16). Weights and context per host
Run it yourself
open weights, NVIDIA Open Model License · fits 6 of the 6 workstations in our guideSelf-host guide ›- Parameters
- 30B
- 3B active per token (MoE)
- Memory needed
- 18 GB
- in INT4 · 36 GB in INT8
- Smallest cloud setup
- RTX 3090
- 24 GB · $0.17/h, about $124 a month
- Worth it above
- 496M tokens
- a month, against the cheapest API at $0.25 per M
| Quantization | Memory | Smallest cloud GPU | Quality |
|---|---|---|---|
| INT4 (Q4) | 18 GB | RTX 3090 · $0.17/h | smallest footprint, minor quality loss |
| INT8 (Q8) | 36 GB | A100 · $1.19/h | near-lossless |
| FP16 (full) | 72 GB | A100 · $1.19/h | reference quality |
At home: in INT4 it fits a RTX 5090 (32 GB) + host PC (plus a host PC), about $3,500 once, or $0.1332/h over 3 years, electricity aside; it pays for itself after about 14B tokens against the cheapest API.
Memory ≈ parameters × bytes per parameter × 1.2 · cloud price: on-demand rate, running 24/7 · builds: GGUF, AWQ, MLX, GPTQ, Ollama · break-even calculator
| Provider | Input | Output | Cache ↑ | tok/s | First token | Uptime 24h | Checked |
|---|---|---|---|---|---|---|---|
| DeepInfrafp4 (bf16) · 256K ctx | $0.05 | $0.20 | $0.025 | — | — | 64.5% | Oct 8 |
| Crusoefp8 (bf16) · 256K ctx | $0.05 | $0.20 | $0.03 | — | — | 100% | Oct 7 |
| Novita AIfp4 (bf16) · 256K ctx · 32K out | $0.05 | $0.20 | — | — | — | — | Oct 23 |
| Nebiusfp8 (bf16) · 256K ctx | $0.06 | $0.24 | — | 173 | 0.38 s | 98.3% | Oct 23 |
4 of 4 providers · prices per million tokens, read Oct 23, 06:00 UTC · speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · uptime: OpenRouter, last 24 h · under each host: weights precision, context and max output as served, the maker's value in brackets when the host's differs
| Provider | Weights | Context | Max output | Tools | Structured outputs | Reasoning |
|---|---|---|---|---|---|---|
| Crusoe | fp8 | 256K | not stated by the host | — | — | — |
| Novita AI | fp4 | 256K | 32K | yes | no | yes |
| DeepInfra | fp4 | 256K | not stated by the host | — | — | — |
| Nebius | fp8 | 256K | not stated by the host | yes | yes | yes |
Weights, context and max output: read at each host's own catalogue or docs where its terms allow, otherwise as the host declares to OpenRouter; hover a cell for its source and date · tools, structured outputs, reasoning: as declared to OpenRouter · not verified by us
Nemotron 3 Nano 30B A3B spec sheet, field by field, with sources
Nemotron 3 Nano 30B A3B spec sheetmaker-declared
What NVIDIA declares about the model itself — read from its own weights repository and documentation, not from any host. What each endpoint actually serves is in the pricing above: the two can differ, and both are true.
- Context window
- 262,144 tokens
- Parameters
- 31,577,937,344≈ 31.6B
- Active experts
- 6 of 128 active per token
- Weight precision
- bfloat16
- License
- NVIDIA Open Model License — open weights, conditionalcuratedScope: the licence on the published weights, and nothing else. It is not the maker's acceptable-use policy, which is a separate document, and it is not the contract you sign with whichever host you call — hosts set their own terms, which this index does not read.
- Knowledge cutoff
- not published by the maker≈ June 2024 third-party estimate — not the maker's figurecurated
- Modalities
- text
- Max output (official)
- to verify
- Release date
- to verify≈ December 15, 2025 third-party catalogue — not the maker's figure
- Maker lifecycle status
- to verify
Also declared in the weights
- Architecture class
- NemotronHForCausalLM
- Layers
- 52
- Vocabulary
- 131,072 tokens
- Weights first published
- December 4, 2025
- Weights last updated
- August 24, 2026
- Languages declared
- English · Spanish · French · German · Japanese · Italian
- Papers
- arXiv:2512.20848 · arXiv:2512.20856
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 · config.json @ bf77c31 · read 2026-10-08 · 6 of 10 fields published by the maker
| Provider | Chatbot 1,000 in / 500 out | RAG / long context 8,000 in / 500 out | Batch summarize 4,000 in / 1,000 out |
|---|---|---|---|
| Crusoe | $0.15 | $0.50 | $0.40 |
| Novita AI | $0.15 | $0.50 | $0.40 |
| DeepInfra | $0.15 | $0.50 | $0.40 |
| Nebius | $0.18 | $0.60 | $0.48 |
What each host bills for 1,000 requests of a representative job
Your own workload
Other NVIDIA models · all its models, hosts and history ›
- Nemotron 3.5 Super VL7 providers · from $0.50
- Nemotron 3 Super4 providers · from $0.48
- Nemotron 3.5 Lightning4 providers · from $0.18
- Nemotron 3 Ultra3 providers · from $2.70
- Nemotron 3.5 Content Safety1 provider · from $0.40
- Nemotron 3 Nano Omni 30B A3B Reasoning0 provider · from —