BurnerGPU

What is BurnerGPU?

Rent a private GPU by the hour. Pick a model, pay in crypto or by card where available, get a chat window and an API. No account.

What is Private hosting?

Normally your GPU is rented from any host on Vast.ai, which can be a home or small business. With Private hosting, the GPU runs only in a certified professional datacenter that Vast vets for security (audited operators, locked facilities). Your chat reaches it over an encrypted connection pinned to your session, we never store your chats, and the machine is destroyed when your time ends. What it is not: the datacenter still controls the hardware, so the operator could in principle see what runs on it. Full hardware-level encryption (confidential computing) is not available from our supplier yet. It costs more and is only offered where datacenters have supply.

What is BurnerGPU?

You rent a GPU by the hour and run an open AI model on it. You pick a model and a number of hours, pay, and get a private chat window and an OpenAI-compatible API endpoint. When your time is up, the GPU shuts down and the endpoint is deleted.

Who is it for?

People who want to run a 27B to 700B-class open model but do not own the hardware, and do not want a hosted AI that filters answers and keeps their prompts. Developers who want a temporary endpoint for an agent or a test also fit.

What does 'abliterated' mean?

An abliterated model is an open-weights model that has had its refusal behavior removed, so it answers instead of declining. The simple page offers these builds. Stock (unmodified) models are in Advanced. You are responsible for what you ask it. See the rules below.

How do I pay?

In USDC or ETH on Base, or by card where the order page offers it. Each order gets a fresh deposit address. You send the exact amount, the page updates after a few confirmations, and the GPU starts. Prices include tax and are shown in whole dollars before you pay. Card is offered only when the order page shows it; see the card answer below when it appears.

Is this a beta?

Yes. Beta: starts can fail. If yours does, we refund you by hand to the wallet you paid from, within 48 hours. Card orders are refunded automatically.

Do I need an account?

No. There is no signup, no email and no name. Your order page link is your receipt, so bookmark it. Anyone with the link can see the endpoint and key for that order, so do not share it.

How private is it, really?

Crypto orders: we never ask who you are. We do not store your chat. The chat window relays your messages to your GPU and keeps nothing; the conversation lives in your browser tab and is gone when you close it. Honest limits: the GPU is run by a third-party host on a GPU marketplace, and a host can in principle see what runs on its machine. Traffic between us and your GPU is encrypted with a certificate pinned to that machine. The host itself still sees everything that runs on it. Do not treat any of this as protection against a determined, well-resourced adversary.

How do I use it from code or an agent?

When your GPU is live, the order page shows a base URL and an API key. Set OPENAI_BASE_URL and OPENAI_API_KEY and use any OpenAI-compatible client, tool or agent framework. Use the model name shown on the order page. The key works only for your GPU and stops working when the session ends.

Can I chat without writing code?

Yes. Press Open chat on your order page. It is a plain chat window with an optional system prompt. It streams the reply and stops when your time ends.

How long until it starts?

Usually several minutes while a provider is found and the model loads: about 5 to 8 minutes for Qwen 3.8 27B (measured); other models are untested and may take longer. Your clock starts when the endpoint is live, not when you pay.

What happens if it never starts?

Crypto orders: we refund you by hand to the wallet you paid from, within 48 hours. Card orders: the card is refunded automatically. Keep your order id.

What happens when my time runs out?

The GPU is shut down and the endpoint is removed. The rented machine is destroyed, but the host controls the hardware, so save what you need before the timer ends and don't leave anything on it you would mind a host seeing. You cannot extend a session; start a new order.

Why is it priced this way?

The price covers the GPU rental, provisioning, a cushion for startup time, and our margin. It is computed once, shown before you pay, and tax is included. There are no credits and no minimum balance.

Is it fast?

We do not publish speed or uptime numbers because we have not measured them well enough. The GPU is a single rented machine, not a shared API, so speed depends on the model and card.

What is not allowed?

Anything illegal where you live, anything that harms or exploits others, sexual content involving minors, and abuse of the service. We may end sessions and refuse orders. Running an unfiltered model does not move any responsibility from you to us. Read the Terms and Acceptable Use Policy, or what we store. To report abuse, email the address in the footer.

What is it not?

Not an enterprise service: no uptime guarantee, not for regulated or sensitive data, not a replacement for a lawyer or doctor. It is a small early service run by one person. Expect rough edges.

Who runs the GPUs?

Third-party hosts on the Vast.ai GPU marketplace. We rent from them on your behalf and resell by the hour.

Pick a Model