// what is nan
Builders sharing dedicated GPUs to run open models and burn tokens without limits.
Closed-model APIs are convenient but expensive and limited. Open models are free but running them seriously requires hardware almost no one has at home.
NaN solves this by sharing powerful GPUs among builders who actually use them, splitting the real cost of infrastructure.
A closed community by design, limited to the real capacity of shared GPUs. When it fills up, we open a waitlist.
The community lives on Discord. The website only manages access. Every member pays to be here. Only those who truly want to build make it in.
// inference cluster
Inference cluster for open AI models
A shared cluster with NVIDIA RTX PRO 6000 Blackwell, 96 GB VRAM, 256 GB DDR5 RAM and 48 dedicated threads per node. Four open models running at once, chosen by the community each quarter.
hardware
- GPU
- NVIDIA RTX PRO 6000 Blackwell
- VRAM
- 96 GB GDDR7 ECC
- RAM
- 256 GB DDR5 ECC
- CPU
- 48 threads · Intel Xeon Gold 5412U
- Storage
- 2× 960 GB NVMe RAID 1
stack
- Inference
- vLLM / SGLang
- Proxy
- LiteLLM
- API
- OpenAI-compatible
// current models
qwen3.6
llmLLM · 35B-A3B MoE · FP8 · 128K context · Tool calling · Reasoning
qwen3-embedding
embeddingEmbeddings · 8B · 4096 dims · 100+ languages · MMTEB 70.58
kokoro
ttsTTS · 82M params · 67 voices · <1s latency · ef_dora · em_alex
whisper
sttSTT · large-v3 · CTranslate2 · ~3.2% WER Spanish · auto language detection
- Your code doesn't train any models
- Data in EU, no prompt logs
- No token caps
- Personal API key for each member
// privacy
Pure inference. Zero logs.
NaN is inference infrastructure. There is no logging of prompts, model responses, nothing. Everything lives in your local session, on whatever client, agent, or custom app you use.
The API is pure inference. Nothing you do with it is logged except server-side metrics (tokens/s, RPM, etc.) for cluster maintenance and monitoring only.
Your code doesn't train any models. Your prompts aren't stored. Your data doesn't leave your session.
- Zero logs: no prompt or response logging
- Everything lives in your local client or agent session
- Processing in the European Union
- Your code doesn't train any models
- Server metrics only for cluster maintenance (tokens/s, RPM)
// how models are chosen
Models chosen by the community. Refreshed every quarter.
- Every 3 months, members vote on the new models that will run on the cluster.
- Candidates are pre-selected technically: they must fit in 96 GB of VRAM.
- Voting happens on Discord, open to all members.
- The open models ecosystem moves fast. So does NaN.
// current models
qwen3.6
LLM · 35B-A3B MoE
qwen3-embedding
Embeddings · 8B
kokoro
TTS · 67 voices
whisper
STT · large-v3
// who's behind
The community is run by one person. Not a company.
Cristian Córdova
@barckcodeTenerife, Islas Canarias
Founder of Helmcode, working with startups on cloud infrastructure, AI tooling and open source.
Thanks to Helmcode we offer open model inference with unlimited token plans for enterprises, I share my experiments on social media and AI research.
I started NaN because I wanted to run serious open models without usage limits, and share and learn with people in the same boat: builders who actually burn tokens.
// pricing
One tier per region. No upsells. No trial.
NaN Member EU
🇪🇺VAT included
Limited spots
- Access to shared cluster (96 GB VRAM)
- Open models chosen by the community
- Personal API key compatible with OpenAI
- No token caps
- Private Discord channels for members only
- Voice in quarterly model voting
- Month to month, no commitment
Limited spots. When it fills up, waitlist.
NaN Member USA/Latam
🇺🇸Taxes included
Limited spots
- Access to shared cluster (96 GB VRAM)
- Open models chosen by the community
- Personal API key compatible with OpenAI
- No token caps
- Private Discord channels for members only
- Voice in quarterly model voting
- Month to month, no commitment
Limited spots. When it fills up, waitlist.
// join
A community of builders. Limited spots. Shared GPUs with unlimited usage.
We share dedicated GPUs to run open models and burn tokens without limits. Leave your email and we'll reach out when we open payments.
// faq
What you'll probably ask.
- NaN is a closed community of builders who share access to dedicated GPUs to run open-source AI models. Every member pays to be here, and there's only room for those who will actually build.
- Each community member shares a dedicated GPU (RTX PRO 6000 Blackwell, 96 GB VRAM). The GPUs run open models 24/7 on a shared cluster. Each member gets a personal OpenAI-compatible API key to use models as they want, with no token caps.
- That's right. Once you have cluster access, you can burn tokens without limits. The models you run depend on what the community votes to run each quarter. No quotas, no throttling, no surprises on the bill.
- Every quarter, community members vote on which new models will run on the cluster. Candidates are pre-selected technically to ensure they fit within 96 GB of VRAM. Voting happens on Discord, open to all members.
- Data is processed in the European Union, with no prompt logging. Your API key is personal and non-transferable. We don't share data between members or sell information to third parties. Your code doesn't train any models.
- For builders who want to run serious AI models without the overhead of managing infrastructure. If you're a startup, researcher, or maker who needs inference at scale without token limits, NaN is for you. It's not for casual experimentation.
- Because we believe real innovation comes from tools you can audit, modify, and use without restrictions. Closed models are convenient but create dependency. With open models you have full control over your stack.
- No. NaN is for inference, not training. The GPUs are optimized to run existing models at high speed, not for fine-tuning. If you need training, there are other platforms for that.
- We currently accept credit/debit cards and cryptocurrency. Prices include taxes (VAT for EU, taxes for USA/Latam). You can cancel anytime, no commitment.
- Leave your email on the waitlist and we'll reach out when we open payments. The community lives on Discord — once you join, you get access to all private channels, quarterly votes, and direct support from the team.