Llama 4 Scout with long context
Meta Llama 4 Scout — MoE with ultra-long context
- Video memory
- 200 GB
- Setup
- ~15 min
- Access
- port 8000
- Billing
- hourly, in BRL
Scout is Meta's variant built for very long context: whole documents, repositories, complete histories. It needs access approved on the model repository before the first download.
Meta's Llama 4 Scout served via vLLM with an OpenAI-compatible API — MoE with a huge context window (up to ~10M tokens). IMPORTANT: GATED on Hugging Face → set HF_TOKEN with approved access. Large model: prefer multiple H100/H200 GPUs; long context uses a lot of VRAM.
What it is for
- Analyse a full contract or case file without chunking
- Agents with long conversational memory
- Search and summarisation over a large document base
- The Llama ecosystem, with plenty of compatible tooling
How to deploy
- Create your account and add balance (card or Pix, no subscription).
- In the console, pick the Llama 4 Scout 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
Do I need to request access?
Yes, it is a gated model. Accept the terms on your repository account and provide the token in the template.
How much video memory?
Around 200 GB, meaning a block of large cards.
Can I use maximum context right away?
Very long context costs extra memory. Start smaller and raise it as the card allows.