Every GPU. One API. Priced by the market.
Bring idle GPUs and cloud API keys to the marketplace — or tap into hundreds of independent providers through one unified endpoint.
4.9on G2
4.7on Capterra
4.8on TrustRadius
Llama
Claude
GPT
Mistral
DeepSeek
Qwen
GeminiHow it works
Two sides. One marketplace.
Providers list OpenAI-compatible capacity. Developers top up Air Credits and call one Air model ID. Air always uses that provider and meters exact PAYG usage.
Get an air_ API key
Sign up, finish onboarding, and copy your key once. Point the OpenAI SDK base URL at Air.
Enable models + buy Credits
Open a marketplace listing, enable the models you need, then top up Air Credits.
Call /api/v1
Send one Air model ID per Chat Completions request. Usage is charged from Credits.
For providers
List your compute. Earn from idle capacity.
Create a listing, add model offers, validate your endpoint, then go live. Developers enable your Air model IDs and pay with Credits.
OpenAI-compatible endpoints
vLLM, llama.cpp server, RunPod wrappers — anything that speaks /chat/completions.
List any capacity
One listing can include multiple models. Each gets its own Air ID and PAYG rates.
You set the rates
Compete on $/M tokens. Developers enable models on your listing page.
Usage & earnings
Track routed traffic and accrued net from the dashboard. Connect payouts come later.
Provider preview
Live listing
PAYG rates · Credits · routed /api/v1 traffic
Requests
2.4M
Fill rate
94%
Avg $/M tok
$0.41
1import OpenAI from "openai";2 3const client = new OpenAI({4 apiKey: process.env.AIR_API_KEY,5 baseURL: "https://www.airinference.com/api/v1",6});7 8const completion = await client.chat.completions.create({9 model: "air_gpt4o_pa_7K4D",10 messages: [{ role: "user", content: "Hello, Air." }],11});For developers
One API. Every model. Marketplace pricing.
Point your SDK at Air with baseURL. Open a listing, enable models, top up Credits, then call each Air model ID.
- OpenAI-compatible chat completions
- Browse listings, enable models you need
- One Air model ID per request
Marketplace pulse
0+
Active Providers
0+
Models Available
0M+
Requests served
0%
Avg. Savings vs Direct
Live listings
Models priced by competing hosts
Rates tick as providers compete. Same model family, different hosts — pick one or let the router choose.
Llama 3.3 70B
Claude 3.5
GPT-4o class
Mistral Large
DeepSeek V3
Qwen 2.5 72B
Gemini Flash
Command R+
Pricing
Competition beats fixed rate cards
Single-platform pricing locks you to one inventory. Marketplace hosts compete — rates move with supply.
From the marketplace
What builders and operators say
Real workflows from both sides — topping up Air Credits, listing OpenAI-compatible capacity, and calling exact Air model IDs with one API key.
Developer
“We pointed our OpenAI SDK at Air, enabled models on a listing, topped up Credits, and shipped the same night.”
Provider
“Two idle H100s were sitting cold after a cancelled contract. Listed an OpenAI-compatible endpoint Friday afternoon; by Tuesday PAYG Credits traffic was covering colo power for the month.”
Developer
“Pay-as-you-go Credits are boring in a good way — we see exact spend per request. Spending limits saved us from a silent overage when an agent loop got chatty.”
Provider
“Transparent $/M rates are where the real volume shows up. A buyer enabled our model, topped up Credits, and traffic hit our vLLM box within an hour.”
Developer
“Enabling a specific Air model ID for latency-sensitive work is the split we wanted. One air_ key, one provider per ID.”
Provider
“Credentials stay encrypted and buyers never see our upstream key. Going live without that would have been a non-starter for our compliance review.”
Early access
Join the compute marketplace
Whether you have spare capacity or need market-priced inference — create an account. Dual-path from day one.