Full DeepSeek-R1, on a multi-GPU block
Full DeepSeek-R1 (671B) — SOTA reasoning
- Video memory
- 700 GB
- Setup
- ~15 min
- Access
- port 8000
- Billing
- hourly, in BRL
The full R1 is a heavyweight reasoning model and needs a block of GPUs to fit in memory. This template brings up vLLM with sharding configured — you only choose the machine.
Full DeepSeek-R1 (671B MoE) served via vLLM with an OpenAI-compatible API. Frontier reasoning (AIME, math, code). NOTE: needs multiple H100/H200 GPUs or a quantized variant (tune tensor-parallel and the model). For a single GPU, use the DeepSeek-R1-Distill template.
What it is for
- Frontier-grade reasoning without sending the problem outside
- Internal evaluation of an open frontier model
- Research and comparison under a permissive license
- Heavy usage at a predictable hourly cost
How to deploy
- Create your account and add balance (card or Pix, no subscription).
- In the console, pick the DeepSeek-R1 (full) template and a machine — the console hides the ones that do not meet the requirement.
- In about 15 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
How much video memory?
On the order of 700 GB — a block of eight large cards. Check availability in the console before reserving.
Is there a cheaper option?
Yes, the distilled versions run on a single 24 GB card with much of the result.
How long until it answers?
Weight download is long. Plan a window of several minutes before the first test.