Meta Model API
first_party
Getting credentials
api-key
api_key via header (header Authorization: Bearer)
Environment: MODEL_API_KEY
Create keys in the dev.meta.ai dashboard (API Keys tab). For Claude Code: ANTHROPIC_BASE_URL=https://api.meta.ai with ANTHROPIC_AUTH_TOKEN=<key>.
Endpoints
| ID | URL | Protocol | Auth |
|---|---|---|---|
| v1-responses | https://api.meta.ai/v1/responses | openai-responses | any |
| v1-chat-completions | https://api.meta.ai/v1/chat/completions | openai-chat | any |
| v1-messages | https://api.meta.ai/v1/messages | anthropic-messages | any |
API surfaces
textstreamingfiles
Quirks
- The Llama API (api.llama.com/compat/v1) was retired 2026-07-06 — there is no first-party hosted Llama inference; Llama models are open weights served by third parties. (docs)
- reasoning_effort none returns 400 (always-on reasoning). Chat Completions exposes reasoning_content redacted to empty for external callers; use the Responses surface for summaries and encrypted reasoning replay. No logprobs, no n>1, no stop. (docs)
- All keys share team limits: Standard 3,000 RPM / 4M TPM. Contributor tier (muse-spark-1.2-contributor) trades a heavy discount for prompts possibly training future Meta models. (docs)
Models offered
| Model | Wire ID | Endpoint | Status | Reasoning |
|---|---|---|---|---|
| meta/muse-spark-1-2 | muse-spark-1.2 | v1-chat-completions | ga | effort · default on · mandatory · returns hidden |
| meta/muse-spark-1-2 | muse-spark-1.2-contributor | v1-responses | ga | effort · default on · mandatory · returns reasoning_summary · round-trips encrypted_content |
| meta/muse-spark-1-2 | muse-spark-1.2 | v1-responses | ga | effort · default on · mandatory · returns reasoning_summary · round-trips encrypted_content |