/// quick guide

From zero to a live endpoint in four steps.

No Dockerfiles, no driver installs, no cluster to babysit. Here is the entire path from picking a model to your first completion.

1

Pick a model

Browse the catalog and choose a model that fits your task — chat, reasoning, vision, or code. Each card shows its precision and minimum GPU up front.

Browse models
2

Choose a GPU tier

Every model lists small, medium, and large tiers with an exact instance type and hourly price. Pick the tier that matches your context length and concurrency.

3

Deploy

Hit deploy, choose a region, and set an optional TTL. GPU Router provisions the dedicated instance, pulls the weights, and starts vLLM for you.

gpurouter CLI
# or do it all from the terminal
gpurouter deploy glm-4-9b-chat \
  --tier medium \
  --region us-east-1 \
  --ttl 2h
4

Send a request

Once the endpoint is running you get an OpenAI-compatible URL and an API key. Point any OpenAI client at it — no code rewrite needed.

python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.gpurouter.dev/v1",
    api_key="grk_live_...",
)

resp = client.chat.completions.create(
    model="glm-4-9b-chat",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)

Tips for smooth sailing

  • Always set a TTL on frontier deployments (like GLM-5.2) so an 8× H200 node never runs longer than you intend.
  • Start on the small tier to validate your prompt, then scale to medium or large for production context lengths.
  • FP8 models require an Ada or Hopper GPU. GPU Router blocks incompatible tiers automatically, so you cannot misconfigure.
  • Deploy in the same region as your weights bucket for fast, free in-region transfer.

Ready to deploy your first model?

Create an accountBrowse the catalog