Instant access to 1,000+ NVIDIA GPUs and to AI models by the token

Billing is hourly and in Brazilian reais (R$), with live prices shown on the site — from entry-level RTX cards to NVIDIA H200, B200 and B300. You only pay for the time you use, with no contract.

GPU instances are deployed in seconds. After payment, your GPU is ready to use immediately with SSH access.

We offer NVIDIA H200, H100, A100, L40, RTX 6000, RTX 4090, RTX 4080, RTX 3090, and more. All GPUs come with dedicated vCPU, RAM, and storage.

Yes. The API is OpenAI-compatible: point the base URL at https://gpubrazil.com/v1 and use your gpub_live_ key. Any SDK or tool that already speaks to OpenAI works without a code change. Billing is per token, in reais, from the same balance that pays for GPUs — no subscription, no minimum, and none of the Brazilian IOF tax charged on purchases abroad.

Yes. The endpoint is POST /v1/images/generations, in the same OpenAI format and with the same key. There are three image generation models; some are billed per image and others per image token, and each price is shown live on the pricing table. Image models do not answer on the chat endpoint, and vice versa.

Yes, on every model in the catalogue that accepts images — Claude, GPT, Gemini and the open-weight ones. Use the standard OpenAI image_url block inside content, with no change to your code. Audio input also works on the models that support it, through the input_audio block. If a model does not accept what you sent, the API refuses with an error explaining why instead of silently dropping the part it cannot read.

Renting a GPU gives you the whole machine, with root and SSH access: you pick the model, control the weights, the logs and the data, and pay by the hour while it runs — that is the route for training, fine-tuning, video generation or serving your own model. The per-token API needs no machine at all: you call the model and pay only for the tokens of each call. Plenty of people use both, from the same balance.