Meta

It does NOT sell any Llama model per token. The older Llama-branded developer API host (llama.developer.meta.com) now 302-redirects to ai.developer.meta.com, and www.llama.com 301-redirects to developer.meta.com/ai/ — the Llama-chat-model API is gone as a first-party product. Every "meta-llama/..." Hugging Face repo is an open-weight download, not a Meta-priced API endpoint; any per-token price attached to those IDs comes from a third-party host (Together, Fireworks, Groq, Bedrock, Vertex, OpenRouter, etc.), never from Meta, so all Llama and Muse Glimmer entries here carry -1 prices by design.

Meta's open-weight line now has two branches: the Llama Community-License models (Llama 4, Llama 3.x) and Muse Glimmer, a 30B model distilled from Muse Spark and released under Apache 2.0 — Meta's first genuinely permissive open-weight LLM licence. Muse Code is Meta's terminal/CI coding agent built on Muse Spark; it bills against the same Model API tokens or a flat monthly subscription. Meta publishes no batch-API discount, no long-context premium ("you pay the same rate whether your context window is mostly empty or almost full"), and no knowledge-cutoff or release date for any Muse model.

A Access and limits

Self-serve. Create a key in the Model API dashboard and pass it as a Bearer token (Authorization: Bearer $MODEL_API_KEY; keys look like LLM|<id>|<secret>). Drop-in compatible with the OpenAI SDK, the Anthropic SDK, and OpenAI-compatible agent CLIs — three protocol surfaces are exposed on the same models and the same price: Responses (/v1/responses), Chat Completions (/v1/chat/completions) and Anthropic-style Messages (/v1/messages).

Rate limits are per team, not per key: Standard tier 3,000 RPM / 4,000,000 TPM; Contributor tier 100 RPM / 3,000,000 TPM; Muse Image 150 RPM with no TPM cap; Muse Voice Transcribe is metered by audio, capped at 8 concurrent streams and 1,000 streams/hour. Background responses have a separate 600 submissions/minute/team cap. Two extra billable items sit outside the per-token table: web search grounding at $2.50 per 1,000 search queries on text models, and nothing extra for Muse Image's built-in search (included in the per-image price).

Meta injects steering context into every prompt that is explicitly not billed. Choosing a -contributor model ID is a data-licensing decision, not just a discount: Contributor tier grants Meta permission to train on your prompts and completions, while Standard tier does not. Open-weight models (Llama 4, Llama 3.x, Muse Glimmer) are downloads from Hugging Face — no API key, no gated access for Muse Glimmer — and you pay only your own compute or a third-party host.

C Calling the API

Meta accepts requests in the OpenAI Chat Completions format, so any OpenAI-compatible client works by changing the base URL. Example uses muse-spark-1.3-contributor.

curl https://api.meta.ai/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "muse-spark-1.3-contributor",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

from openai import OpenAI

client = OpenAI(
    base_url="https://api.meta.ai/v1",
    api_key=os.environ["API_KEY"],
)

response = client.chat.completions.create(
    model="muse-spark-1.3-contributor",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

A compatibility layer is not always the provider's full API — caching, strict tool schemas and structured output are commonly unsupported there. Check the official docs before relying on an advanced feature.

D Official references

• Homepage
• API documentation
• Pricing