Available on request
Llama API
for Meta’s open-weight models.
Llama 3.3, Llama 3.1 and Llama 3.2 Vision cover most of what teams ask Meta’s models to do. If you want one of them behind an OpenAI-compatible API, tell us which and how much you’ll use, and we’ll talk through adding it to the Xinference Model API.
Pick a Llama
Three Llama lines, three jobs
Llama 3.3 70B
A 70B instruction model for chat and tool use. The usual default when you want one capable general model.
Llama 3.1: 8B, 70B or 405B
The same generation in three sizes. 8B for fast, cheap calls; 405B when you need the most capacity Meta published in this line.
Llama 3.2 Vision: 11B or 90B
Built for visual recognition, image reasoning, captioning and questions about an image.
Shared across the family
Context and languages
| Context window | 131,072 tokens for Llama 3.3, 3.1 and 3.2 Vision Model maker’s figure |
| Languages | English, German, French, Italian, Portuguese, Hindi, Spanish and Thai |
| Licence | Llama Community License for each version, not Apache 2.0 |
Licence in plain English
Three conditions to know before you build on Llama
| Condition | What it means |
|---|---|
| “Built with Llama” | If you make Llama, or a product that contains it, available to others, the licence asks you to display “Built with Llama” prominently and pass on a copy of the licence. |
| 700 million monthly users | Organisations above 700 million monthly active users at release need a separate licence from Meta. |
| Acceptable Use Policy | Use must follow Meta’s Acceptable Use Policy for Llama and applicable law. |
Sources: Meta’s Llama 3.3 model card and the Llama 3.3 Community License, checked 23 October 2026. This is a summary, not legal advice. Llama, Meta and the Meta logo are trademarks of Meta Platforms, shown only to identify the model. Their use does not imply endorsement.
Llama API questions
Is Llama free for commercial use?
For most companies, yes, under the Llama Community License, with conditions: the “Built with Llama” notice, Meta’s Acceptable Use Policy, and a separate licence for organisations above 700 million monthly active users.
Which Llama model should I use?
Start with Llama 3.3 70B for general chat and tool calls. Drop to Llama 3.1 8B for simple, high-volume tasks, and pick Llama 3.2 Vision when images are part of the input.
Can I get a Llama API on Xinference?
Llama is available on request. The Model API’s standard models are Kimi K3 and DeepSeek V4.1 Flash, with public per-token pricing. For Llama, tell us the version and expected volume.
Does Llama support languages other than English?
Meta lists eight supported languages for these models: English, German, French, Italian, Portuguese, Hindi, Spanish and Thai.
Last updated