One key in front of every way a model can be served.

Any OpenAI or Anthropic SDK, one base URL, the exact cost of every call.

One request is routed to one of four ways of serving a model, and every call is priced on one bill.POST /v1/chat/completions{"model": "…", "messages": […]}route by modelANozzle's keyOpenAI, Anthropic, Gemini…BShared podopen weights, our GPUCYour keybring your own credentialDYour poddedicated GPUevery call priced · by workspace, project, key and request

Point any SDK at one base URL.

The OpenAI-shaped surface lives under /v1, the Anthropic drop-in under /anthropic. Same key either way, and every reply tells you what it cost.

  • Chat, responses, embeddings, rerank, transcription, speech, images, moderation.
  • Streaming on every text surface; prompt caching preserved on the Anthropic path.
  • Capabilities are refused by name, never silently dropped.
  • X-Nozzle-Cost-Micro-Cents on every buffered reply.
curl
curl https://api.opennozzle.com/v1/chat/completions \
  -H "Authorization: Bearer pk_live_…" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-4.1-mini", "messages": [{"role": "user", "content": "Hello"}]}'
python
from openai import OpenAI

client = OpenAI(api_key="pk_live_…", base_url="https://api.opennozzle.com/v1")
client.chat.completions.create(model="claude-haiku-4-5",
                               messages=[{"role": "user", "content": "Hello"}])

Two axes, four modes, one wire.

Whose credential pays the upstream, and who gets the capacity. Move a model between modes chasing price, latency or data residency without touching a line of caller code.

A

Platform passthrough

Nozzle's credential, shared capacity · pay per token

Every closed model without an account to manage.

B

Shared self-hosted

Our pod, shared capacity · pay per token

Open weights served once, called by everyone.

C

Bring your own key

Your credential, shared capacity · pay platform fee

When the tokens must land on your invoice.

D

Dedicated self-hosted

Your pod, dedicated capacity · pay per GPU-hour

A machine nobody else touches.

A subscription, and compute at cost.

Nozzle does not mark up GPU time. You pay for platform access; tokens and GPU-hours pass through at the cheapest rate we can source.

Full pricing →