Updated

Meta Model API

Meta's pay-as-you-go API for its Muse Spark models, plus image, transcription and segmentation models, with OpenAI- and Anthropic-compatible endpoints.

No free tierPaid from Pay per token

Profile checked

Ask your AI about this, with this page as the source:ChatGPT ↗Claude ↗Perplexity ↗
Recent changesFollow Meta Model API (RSS)
  • Oct 2026 Now covered: Meta Model API, for llm api.

How makers use Meta Model API

What Meta Model API is used for

LLM API · Model provider →Meta's Muse Spark models for agents and coding, with a 1M-token context, multimodal input and OpenAI- and Anthropic-format endpoints.Trade-off: Closed weights and a young API (public preview from July 2026); the cheap contributor tier gives Meta the right to train on your prompts and outputs.

Meta Model API vs the alternatives makers chose

OptionMakers' products2026 launchesFree forAt 10,000 usersOpen source
OpenAI API736 80%77%~1,000 users$10/mo · GPT-6 LunaNo
Claude252 27%21%~1,000 users$10/mo · Claude Haiku 5.5No
Gemini API99 11%12%~360 users$28/mo · Gemini 3.1 Flash-LiteNo
Groq39 4%3%~1,500 users$6.75/mo · GPT OSS 20BNo
Azure OpenAI22 2%0%——No
Mistral AI15 2%1%~1,700 users$6.00/mo · Ministral 3 (3B)No
Meta Model API0 0%0%~95 users—Unknown

For LLM API: how many of the 923 makers' products we know the LLM API of chose each option, and the share among the 256 launched in 2026, since a tool that came first has products from every year, and what each costs at 10,000 users.

Costs are the cheapest plan that fits, from each vendor's pricing page, at the usage our stack planner assumes for a product with that many users. Free for: the users its cheapest plan a business can use stays free for.

When to use Meta Model API, and when not

Something else fits better if

  • You want the strongest general models and the widest ecosystem of examples and integrations: OpenAI API or Claude or Gemini API.
  • Token cost dominates your budget, for example high-volume batch or agent workloads: DeepSeek or Groq or Mistral AI.
  • You want to try or mix many models with one key and one bill, and fall back when a provider is down: OpenRouter.
  • You want that routing and spend control but self-hosted, with keys and logs on your servers: LiteLLM.
From each decision's guide, written from what makers' products chose and why.

Meta Model API is Meta's paid developer API for the models from its superintelligence lab. Its main model is Muse Spark — currently version 1.3, with 1.2 and 1.1 still served — a closed-weight model aimed at agents and coding, with a 1M-token context window and text, image, video, PDF and audio input. The same account also covers Muse Image for generating and editing images, Muse Voice Transcribe for speech to text, and Segment Anything Model 3.1 for finding objects in images and video. It opened in public preview in July 2026; the older open-weight Llama models are not part of it.

More about Meta Model API

There is no Meta-specific SDK. The API speaks the OpenAI Responses and Chat Completions formats at https://api.meta.ai/v1 and the Anthropic Messages format at https://api.meta.ai, so you use the official OpenAI or Anthropic SDKs, or agent tools such as Claude Code, Codex and OpenCode, by changing the base URL and key. You pay per token with no minimums after a team admin adds a payment method.

Useful for a small team: Muse Spark has one price whether the context is nearly empty or full, prompt caching is automatic, and limits are generous from the start (3,000 requests and 4M tokens per minute per team). A contributor tier runs the same models at a small fraction of the price in exchange for letting Meta train on your prompts and outputs — a cheap option for prototypes and internal tools where that is acceptable. A Muse Code subscription is sold separately for Meta's own coding agent.

The main limits: the weights are closed, so you can't host Muse Spark elsewhere (Meta's open-weight model is the smaller Muse Glimmer, which you run yourself), and the API is young, with three Muse Spark versions released since July 2026. Anthropic compatibility is partial — some fields, such as top_k and stop_sequences, are rejected — and the contributor tier's training terms make it unsuitable for customer data.

Who uses it

No maker's live product on record yet.

We haven't found a product that uses Meta Model API yet.

Alternatives to Meta Model API

All alternatives by situation →

Further reading

Questions makers ask about Meta Model API

What are the alternatives to Meta Model API?

For LLM API, the other options makers' products we track chose most are OpenAI API (736), Claude (252), Gemini API (99), Groq (39), Azure OpenAI (22), Mistral AI (15). Each alternative, with the situation it fits, is on the Meta Model API alternatives page.

Can I use the OpenAI or Anthropic SDK?

Yes. Point the OpenAI SDK at https://api.meta.ai/v1 for the Responses and Chat Completions APIs, or the Anthropic SDK at https://api.meta.ai for the Messages API, and pass your Model API key. Note that the OpenAI SDK won't read MODEL_API_KEY on its own, so pass the key explicitly. source ↗

Does Meta train on my prompts?

Not on the standard tier — prompts and completions sent to the standard muse-spark models are not used to train Meta models. The cheaper contributor-tier models (IDs ending in -contributor) are priced lower precisely because Meta may use your prompts and completions to train future models. source ↗

What are the rate limits?

On the standard tier, 3,000 requests and 4M tokens per minute; on the contributor tier, 100 requests and 3M tokens per minute. Limits apply per team, so all API keys in a team share them. source ↗

Does a long prompt cost more per token?

No. There is no long-context premium, so the per-token price is the same up to the full 1M-token window. Repeated prompt prefixes are cached automatically and billed at the lower cached-input rate. source ↗

Do I need to set up prompt caching?

No. Caching is automatic for repeated prompt prefixes, with no key or setup. On the Responses API you can ask for longer retention, up to 24 hours, though the server may still evict entries early. source ↗

Can I run Muse Spark on my own servers?

No. Muse Spark, Muse Image and Muse Voice Transcribe are only served through the API. Meta's open-weight option is Muse Glimmer, a smaller model distilled from Muse Spark under the Apache 2.0 license, which you download and run with tools such as vLLM or llama.cpp. source ↗

Is a Muse Code subscription the same as API access?

No. A Muse Code subscription is a flat monthly plan that works only inside Meta's Muse Code CLI; any other API keys you create in the account are billed pay-as-you-go. source ↗

Is Meta Model API free?

No permanent free tier; paid use starts at Pay per token. source ↗

Can AI coding agents work with Meta Model API?

No llms.txt, official MCP server or CLI found yet.

Who uses Meta Model API?

No tracked product yet. source ↗