Deploy Meta Llama models on dedicated GPUs
5 Meta Llama models ready to deploy on GPU capacity reserved for you, from 279M to 402B parameters, covering chat & reasoning and safety & moderation. Private OpenAI-compatible endpoint, European GPUs, billed per GPU-hour.
| Model | Type | Params | Context | License |
|---|---|---|---|---|
| Meta Llama | Chat & reasoning | 71B | 128K | Llama 3.3 Community License |
| Meta Llama | Safety & moderation | 279M | — | Llama 4 Community License |
| Meta Llama | Chat & reasoning | 109B | 1.3M | Llama 4 Community License |
| Meta Llama | Chat & reasoning | 402B | 512K | Llama 4 Community License |
Llama-Guard-4-12Bvision Meta Llama | Safety & moderation | 12B | 160K | Llama 4 Community License |
Specs from Hugging Face, synced 2026-10-27. Machine-readable version: https://sference.com/models/meta-llama.md
FAQ
Which Meta Llama models can I deploy on sference?
5 today, most requested first: Llama-3.3-70B-Instruct, Llama-Prompt-Guard-2-86M, Llama-4-Scout-17B-16E-Instruct, Llama-4-Maverick-17B-128E-Instruct and Llama-Guard-4-12B. Any other Meta Llama checkpoint or a fine-tune of one can be requested with the form on this page.
How many GPUs do Meta Llama models need?
It depends on the model: the lineup runs from 279M to 402B parameters. Each model page lists the checkpoint size and the minimum footprint on 80 GB and 141 GB GPUs. You don't pick hardware on sference: you choose a performance tier and we size the deployment.
Can I use Meta Llama models commercially?
0 of the 5 listed models are under licenses that allow commercial use (Llama 3.3 Community License and Llama 4 Community License across the lineup). Each model page summarizes its license and links the full text.
How is a dedicated deployment priced?
Per GPU-hour on a monthly commitment, sized to your throughput and latency targets.
Request a Meta Llama model
Tell us which Meta Llama model and roughly how much traffic you expect. Fine-tunes of these models work the same way.
- 01We size it
A performance tier, the capacity behind it, and a quote for your volume.
- 02We benchmark it
Bring one workload with a sample of real traffic; we come back with latency and throughput numbers.
- 03It goes live
We provision the deployment. You call it with the same API and model id as before.