Meta Model API

What it does

Meta Model API is a hosted inference API for Meta's models. Text models (Muse Wonder and Muse Spark) take prompts, files, images and video and return text, tool calls and structured JSON. Muse Image generates and edits images from text prompts. Muse Voice Transcribe turns streaming or file audio into transcripts with speaker labels. SAM 3.1 returns boxes and masks for objects named in a short text prompt.

Developers call it over the Responses, Chat Completions or Messages formats, or through OpenAI- and Anthropic-compatible SDKs and coding agents, keeping existing code and changing the base URL and key. Text models offer 1M-token context, parallel tool calls and reasoning that carries across turns. A discounted Contributor mode on some Muse Spark models trades lower prices for permission to train on your content.

It suits developers building apps, agents, voice or vision features. Access needs an API key sent as a bearer token. Muse Wonder is in preview and has no Contributor mode. Open-weight models (Muse Spark Open, Muse Glimmer) are self-hosted and not served by this API. Muse Voice Transcribe gives turn-level timestamps only, not word-level, and does not synthesize speech.

What it can do

• Text generation and long-horizon agent work with Muse Wonder and Muse Spark (1M-token context)
• Parallel and streamed tool calls, structured JSON output, and search grounding
• Image generation and multi-turn editing with Muse Image
• Streaming and file speech-to-text with speaker attribution via Muse Voice Transcribe
• Object segmentation in images and video with SAM 3.1

Works with

• OpenAI SDK
• Anthropic SDK
• OpenCode
• Codex
• Claude Code

Example workflows

Add live transcription to a call tool

Input: live audio stream. Steps: open the realtime endpoint with your key, send audio, enable speaker labels and keyword biasing. Output: transcript text with turn-level timing and speaker labels.

Build an agent on an existing SDK

Input: an OpenAI-compatible client and tool definitions. Steps: change the base URL to the Model API, set the model to a Muse Spark or Muse Wonder name, run the tool-calling loop. Output: model responses and tool calls.

Before you choose

• Contributor mode lowers token prices but allows Meta to train on your prompts and outputs; Standard mode does not
• Rate limits apply per team across all keys, and each model has its own limits
• Muse Wonder is in preview

Listed under

• Coding
• Coding agents
• Sandboxes and infrastructure
• AI models
• API
• Coding agent
• Developer API
• Developers
• Fine-tuning

Get started

Set MODEL_API_KEY and follow the quickstart

1. Obtain a Model API key and set it as MODEL_API_KEY
2. Point your OpenAI or Anthropic SDK at https://api.meta.ai/v1
3. Send a first request following the quickstart, choosing a model name

Muse Wonder Standard: $4.00 per 1M input, $20.00 per 1M output tokens, cached input $0.20. Muse Spark rates are on the rate table. Muse Code subscriptions are separate and not part of API pricing.

Source: dev.meta.ai/docs/pricing-rate-limits, checked 28 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Get started with Meta Model API and Muse Code - Muse Wonder, Muse Spark, Muse Image, Muse Voice Transcribe, Segment Anything Model, Muse Spark Open, and Muse Glimmer - Meta Model API ↗dev.meta.ai
• Pricing and rate limits - Meta Model API ↗dev.meta.ai

Company website

dev.meta.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.