Llama (Meta) API: endpoints, authentication, models and limits
Connection details
| Base URL | varies by host |
| Auth | Bearer token via LLAMA_API_KEY |
| OpenAI SDK compatible | Yes |
| SDKs | OpenAI SDK against any host, provider-specific SDKs |
Models and token pricing
| Model | Model ID | Context | In / 1M | Out / 1M |
|---|---|---|---|---|
| Llama 4 Maverick | llama-4-maverick-17b-128e | 1000K | $0.200 | $0.600 |
| Llama 4 Scout | llama-4-scout-17b-16e | 10000K | $0.150 | $0.400 |
What is distinctive about this API
- Llama is a set of open weights rather than a single hosted service, so "the Llama API" in practice means whichever host you route to — Together, Fireworks, Groq, Bedrock, or your own vLLM deployment.
- That portability is the point: the same model ID can be served from several vendors, which makes price and latency competitive in a way single-vendor models are not.
- Self-hosting is a first-class option, and the licence permits commercial use subject to its scale threshold.
Constraints worth knowing first
- Quality varies between hosts running the same nominal model, because quantisation and serving configuration differ — benchmark your own prompts per host rather than assuming parity.
- Model naming is inconsistent across hosts; the same weights appear under several IDs.
Frequently asked
- What is the Llama (Meta) API base URL?
- varies by host. Authenticate with a bearer token, conventionally read from the LLAMA_API_KEY environment variable.
- Is the Llama (Meta) API compatible with the OpenAI SDK?
- Yes. The chat completions surface is wire-compatible, so the official OpenAI client libraries work by pointing base_url at varies by host and supplying a LLAMA_API_KEY. No client library change is needed.
- How do I get a Llama (Meta) API key?
- Create an account with the provider and generate a key from the developer console, then expose it to your application as LLAMA_API_KEY. Keys are secrets: keep them server-side, never in client bundles or committed config.
- Which models does the Llama (Meta) API expose?
- Llama 4 Maverick, Llama 4 Scout. Context windows run to 10000K tokens.
Other provider APIs
Meta publishes official documentation at https://www.llama.com/docs. Pricing and limits change; verify against the provider before committing to a budget.