Call hosted models through OpenAI or Anthropic request formats, or launch dedicated GPU VMs.
The open network for inference at scale.
A global mesh of GPUs, all in one network.
Traffic reroutes in milliseconds.
Requests spread across every available node.
Committed uptime, latency, and throughput on dedicated capacity.
OpenAI-compatible. No code changes.
1client = OpenAI(2base_url="https://inference.openrelay.inc/v1",3api_key="sk-••••••••••••"4)
Rent dedicated GPUs or call any model through our OpenAI-compatible API.
GPU compute
Pick a GPU type, VRAM, region, and hourly price from available cards across the network.
Choose an image (Ubuntu 24.04), attach a persistent volume, and launch the runtime.
SSH in with full control, install your stack, then monitor and scale as you go.
Inference endpoint
Use a hosted open model — no node setup, no cold-start wrangling.
Swap the base URL and keep your existing OpenAI-compatible client and tools.
Send requests; routing, autoscaling, and failover are handled for you.
base_url="https://inference.openrelay.inc/v1"Dedicated NVIDIA GPUs with SSH access and a persistent volume. Deposit $5 to get $10 and start deploying.
$0.29/hr
24 GB VRAM
Deploy RTX 4090
$0.39/hr
32 GB VRAM
Deploy RTX 5090
from$0.80/hr
$2.60/hr
80 GB VRAM
Deploy H100Up to 90% cheaper than AWS, Azure, and GCP.
API & SDK
Send chat requests to the inference gateway. Use the model ID from the catalog and your organization API key.
Use OpenAI Chat Completions or Anthropic Messages with supported models.
The gateway authenticates, routes, streams, and records usage for each request.
# Call GLM 5.2 with Chat Completions$ curl https://inference.openrelay.inc/v1/chat/completions \-H "Authorization: Bearer $OPENRELAY_API_KEY" \-H "Content-Type: application/json" \-d '{"model":"openrelay/glm-5.2","messages":[{"role":"user","content":"Explain this stack trace."}]}'# Response{"object": "chat.completion","model": "openrelay/glm-5.2","choices": [...]}
Ready to build?
Get your API key and start deploying in minutes.
export OPENRELAY_API_KEY=vl_••••••••••••••FAQ
Provisioning time depends on the hardware and image you choose. The dashboard reports each stage until your VM is ready. Hosted models need no deployment.
VM workloads run as standard containers. For hosted inference, OpenRelay supports OpenAI Chat Completions and Anthropic Messages request formats.
GPU VMs accrue usage at the displayed hourly rate while they run. Hosted inference is metered by input and output tokens.
Hosted inference routes new requests across healthy backends. VMs are not automatically migrated; failed capacity must be restarted or replaced.
New users receive $1 in welcome credit. Continued usage requires a card and deposit. A first deposit of $5 or more receives a one-time $5 match.
You can apply to become a provider. Approved providers enroll supported GPU or CPU nodes and can track usage and earnings in the dashboard.
An OpenAI-compatible endpoint and GPUs from $0.18/hr. No contract, no minimums — deposit $5 to get $10 and start deploying.