The marketplace for AI models and GPU compute
Call open models through one API, rent GPU hours for fine-tuning, or earn from the GPUs you already own. Priced in rupees, metered per token and per GPU-hour.
- From PKR 0.03
- per 1K tokens
- 6 models
- in the launch catalog
- NVIDIA + Ascend
- GPU and NPU capacity
- OpenAI-compatible
- drop-in API
Launch catalog
Presale pricing on models and GPU capacity.
Multilingual instruction model covering 29+ languages including Urdu. Bilingual support chat, document summaries and form-filling help.
- Params
- 1.54B
- Context
- 32K
- Licence
- Apache-2.0
Small, fast model for intent routing, ticket triage, short JSON extraction and guard-rail checks in front of larger models.
- Params
- 0.49B
- Context
- 32K
- Licence
- Apache-2.0
Hybrid conv/attention architecture tuned for fast inference. Structured extraction, RAG answer synthesis and tool-calling agents.
- Params
- 350M
- Context
- 32K
- Licence
- LFM Open License v1.0
Change one line. Keep your code.
n8bee speaks the OpenAI API. Point your SDK at n8bee, pick a model from the catalog, and every token is metered against your credit.
- Urdu and English models side by side
- Per-key rate limits and usage you can export for finance
- Cost estimated before every fine-tune runs
from openai import OpenAI
client = OpenAI(
base_url="https://api.n8bee.com/v1",
api_key="n8b_live_...",
)
reply = client.chat.completions.create(
model="qwen2.5-1.5b-instruct",
messages=[{"role": "user", "content": "میرے بل کی آخری تاریخ کیا ہے؟"}],
)
print(reply.choices[0].message.content)How it works
Describe the job
Pick a model or declare an intent: fine-tune, serve or evaluate.
We profile it
Memory, GPU-hours and cost are estimated before anything runs.
It runs on matched capacity
Placed on the cheapest pool that meets your device and policy needs.
Pay for what you used
Tokens and GPU-hours are metered per request and drawn from credit.
Presale credit packs
Reserve now, pay at launch, and keep the bonus. Credit works across models and GPU hours.
- 20% bonus credit
- All presale models
- Community support
- 30% bonus credit
- Priority GPU-hour queue
- Private model listing
- Email support
- 40% bonus credit
- Reserved capacity windows
- Dedicated onboarding
- Usage reports for finance
Questions
What am I reserving in the presale?+
A credit pack at a bonus rate. Sign up now and we'll confirm your tier by email before launch. Your bonus is locked in when you pay at launch — nothing is charged today.
Is the API compatible with my existing code?+
Yes. Inference uses the OpenAI chat-completions format, so the official OpenAI SDKs and most frameworks work by changing the base URL and key.
Who runs the GPUs?+
Two kinds of capacity: n8bee-operated hardware, and provider-run GPUs listed by labs and companies. Every listing is labelled so you always know which one a job ran on.
What about my data?+
Datasets and checkpoints stay in governed storage. GPU workers receive short-lived, job-scoped URLs and never get a copy of your workspace.
Does it support Urdu?+
Urdu is a first-class language on n8bee: bilingual models in the catalog, and Urdu corpora for fine-tuning and evaluation.
How do I list my own GPU?+
Join the provider presale on the Sell page. You'll get an enrollment token and a runner container; the runner reports your hardware and starts pulling jobs.
Reserve early access
Join the presale list. We'll confirm your tier and onboarding date by email.