cheapinference.dev
Get started
OPENAI COMPATIBLE. COST CONSCIOUS.

Compute more.
Spend less.

The models you love. A lighter bill.
OpenAI-powered inference at 20% below list price, with an API that feels right at home.

Start building Read the docs

Prepaid credit. No subscription. Open to everyone.

THE SAME OPENAI MODELS0.80× THE COST
OpenAI SDK compatibleUsage you can account forBuilt for your whole teamYour keys, kept private
GOOD MODELS. BETTER ECONOMICS.

Your next idea deserves
a little more headroom.

Choose the model that fits your workload. Keep more of your budget for everything else.

Explore 28 models

gpt-6-astra

OpenAI · Text generation

Released Sep 3, 2026
INPUT / 1M TOKENS$8$10 list price
OUTPUT / 1M TOKENS$40$50 list price
922K input limit · tiered rates

gpt-5.6-sol

OpenAI · Text generation

Released Jul 9, 2026
INPUT / 1M TOKENS$3.2$4 list price
OUTPUT / 1M TOKENS$16$20 list price
922K input limit · tiered rates

gpt-5.6-terra

OpenAI · Text generation

Released Jul 9, 2026
INPUT / 1M TOKENS$1.6$2 list price
OUTPUT / 1M TOKENS$9.6$12 list price
922K input limit · tiered rates

Effective cash price per 1M tokens, standard tier. Usage credit is deducted at list rates. Long-context pricing may differ.

FAMILIAR BY DESIGN

A small change.
A better bottom line.

Keep your SDK, your tools, and your workflow. Getting connected takes a new key and a new base URL.

  1. 01
    Make yourself at home

    Create your account and set up a workspace.

  2. 02
    Create your API key

    Add credit and set a monthly limit for your key.

  3. 03
    Build what comes next

    Point your OpenAI SDK at our API. You’re ready.

Follow the quickstart 
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://www.cheapinference.dev/v1",
  apiKey: process.env.CHEAPINFERENCE_API_KEY,
});

const response = await client.responses.create({
  model: "gpt-5.4-mini",
  input: "Build something great.",
  max_output_tokens: 1024,
  store: false,
});

console.log(response.output_text);
Your existing SDK. A new base URL.
THE DETAILS ARE TAKEN CARE OF

A proper home for your inference.

Everything your team needs to stay in control, from the first request to the monthly review.

One team. Shared context.

Separate organizations, shared balances, and clear roles. Invite teammates without sharing credentials.

Every token has a place.

See requests, input and output tokens, cached usage, latency, and cost. Break activity down by model and key.

Keys with boundaries.

Create a key for each integration, set monthly usage limits, and revoke access when plans change.

SIMPLE, ALL THE WAY DOWN

Less on the bill.
More in the balance.

Pay for credit as you need it. No subscription, no minimum monthly commitment, and a clear record of every deduction.

How pricing works
THE CREDIT EQUATION
$80You pay
→
$100Usage credit
20% lower effective cost
Shared across your organization
No monthly subscription
A LITTLE MORE CONTEXT

Good questions. Clear answers.

Explore the documentation
How does the 20% savings work?

You buy prepaid usage credit at a 20% discount: $80 buys $100 of usage value. Requests deduct credit at the model's published OpenAI list rate. The effective cash cost is therefore 20% lower, with no subscription required.

Do I need to change my application?

Keep your OpenAI SDK. Change its base URL and use a cheapinference.dev API key. Chat Completions, Responses, Embeddings, streaming, and supported function tools use familiar request formats. Check the API reference for the exact supported parameters.

Can I use this with my team?

Yes. Organizations share credit and usage visibility. Owners and admins can invite teammates, manage API keys, and handle billing. Members can view the organization's activity. Anyone can create an account and workspace.

What happens to my prompts and responses?

We do not persist prompt or response bodies. Usage records contain metadata such as model, token counts, cost, status, and latency. Requests are processed by OpenAI and remain subject to its applicable data policies.

Which features are supported?

Text generation, embeddings, streaming, client-executed function tools, and image inputs on compatible models are supported. Image generation, audio, built-in tools, stored conversations, background jobs, and batch are not currently supported.

MAKE ROOM FOR WHAT’S NEXT

Great ideas shouldn’t
come with bigger bills.

Your next chapter starts with a single API key.

Start building