DeepSeek-R1 Distill: reasoning without a giant machine
DeepSeek-R1's strong reasoning that fits on a single GPU
- Video memory
- 20 GB
- Setup
- ~6 min
- Access
- port 8000
- Billing
- hourly, in BRL
The distilled R1 variants bring the reasoning chain down to a size that fits an ordinary card. It is the best starting point for trying step-by-step reasoning without renting a block of eight GPUs.
Distilled DeepSeek-R1 (default: Llama-8B) — inherits R1's step-by-step reasoning in a small model that runs on an RTX 4090. Served via vLLM with an OpenAI-compatible API. Switch to the 14B/32B distill (more VRAM) by editing the model.
What it is for
- Solve maths and logic problems with visible steps
- Review code and explain why each fix matters
- Run internal evaluations with no per-token cost
- A base for fine-tuning on your own domain
How to deploy
- Create your account and add balance (card or Pix, no subscription).
- In the console, pick the DeepSeek-R1-Distill template and a machine — the console hides the ones that do not meet the requirement.
- In about 6 minutes the setup finishes and the access address shows up in the panel, on port 8000.
Done? Just destroy the machine and billing stops with it. No contract, no minimum commitment.
FAQ
Which card should I use?
From 20 GB of video memory. A 24 GB card is comfortable.
How far is it from the full R1?
The full model reasons better on hard problems but needs hundreds of GB. The distill delivers much of the result at a fraction of the cost.
Does it show its reasoning?
Yes, the reasoning text comes separately from the final answer and you decide whether to display it.