Ollama with a real GPU behind it
Run DeepSeek, Qwen3, Llama and Mistral with one command
- Video memory
- 8 GB
- Setup
- ~5 min
- Access
- port 11434
- Billing
- hourly, in BRL
Ollama is the fastest way to get from zero to running: one command pulls the model and it is live with a compatible API. On a laptop a mid-sized model crawls; on a rented GPU it answers immediately.
Ollama is the easiest way to serve open-source LLMs: pull and run DeepSeek-R1, Qwen3, Llama 3.x, Mistral, GLM and hundreds of models with 'ollama pull'. OpenAI-compatible API on port 11434. Starts with qwen3:8b.
What it is for
- Try several models in one afternoon with no setup
- Prototype against a compatible API before scaling
- Swap models with one command, same client code
- Keep test data out of third-party services
How to deploy
- Create your account and add balance (card or Pix, no subscription).
- In the console, pick the Ollama template and a machine — the console hides the ones that do not meet the requirement.
- In about 5 minutes the setup finishes and the access address shows up in the panel, on port 11434.
Done? Just destroy the machine and billing stops with it. No contract, no minimum commitment.
FAQ
How is it different from vLLM?
Ollama is unbeatable on ease and great for one user at a time; vLLM gets far more out of the card when many users call at once.
What model size can I run?
Depends on the card: 8 GB runs quantised 7–8B models; 24 GB opens room for quantised 30B.
Does it expose an API?
Yes, both its own endpoints and OpenAI-compatible ones.