Pricing and rate limits

Use our models with pay-as-you-go billing on Meta Model API and in Muse Code, or in Muse Code with a monthly subscription. This page covers pay-as-you-go: Model API bills the reasoning models (Muse Wonder, Muse Spark, and Muse Ember) per token, image generation per image, Muse Voice Transcribe by the hour of audio processed, and Segment Anything Model (SAM) per image or video frame segmented, with no minimums or upfront commitment.

Pricing modes

Some models can be used in two modes. Standard mode uses standard pricing. Contributor mode offers discounted pricing or more usage in exchange for permission to use your content — the prompts, code, files, and context you send, plus the model responses you get back — to train future Meta models.

Standard mode

In Standard mode, you pay standard prices and your content is not used to train Meta models. Every model on this page is available in Standard mode.

Contributor mode

On Model API, Contributor mode runs the same models at heavily discounted token prices, in exchange for permission to use your content to train future Meta models. It lowers the barrier to entry for prototyping, testing integrations, and scaling experiments where it's acceptable for Meta to use your content to train future Meta models.

On Model API, use Contributor mode with Muse Spark 1.3 or 1.2 by calling its Contributor-mode variant: muse-spark-1.3-contributor or muse-spark-1.2-contributor. Contributor mode isn't available for Muse Wonder or Muse Ember on Model API.

In Muse Code, subscriptions default to Contributor mode, and you can turn it on or off in settings; see Contributor mode in Muse Code.

Model pricing

Muse Wonder

Model: muse-wonder-1.0.

Muse Spark

Models: muse-spark-1.3, muse-spark-1.2, and muse-spark-1.1 in Standard mode; muse-spark-1.3-contributor and muse-spark-1.2-contributor in Contributor mode. Muse Spark 1.1 is Standard mode only.

Muse Ember

Model: muse-ember-1.0, in Standard mode only.

Muse Ember has two price tiers, and the input tokens in your API request set the tier, cached tokens included. When a request continues a stored conversation with previous_response_id, its input includes that conversation. A request with up to 200K input tokens is billed at the rates in the first column. A request with more is billed at the long-context rates in full, not only for the tokens past 200K. Your invoice and usage dashboard list the two tiers separately.

Token pricing notes

These apply to the reasoning models above.

• Cached input costs less: When part of your prompt matches a cached prefix, you pay the cached-input rate for those tokens. Check cached_tokens in the response to see how many were served from cache.
• No long-context premium on Muse Wonder and Muse Spark: you pay the same rate whether your context window is mostly empty or almost full. Muse Ember bills requests with more than 200K input tokens at higher rates; see Muse Ember pricing.
• Web search grounding: $2.50 per 1,000 search queries, in addition to the request's token cost. This applies to the web_search tool on the reasoning models, Muse Wonder, Muse Spark, and Muse Ember.

Muse Image

Muse Image is billed at a flat rate per generated image. The price is the same regardless of prompt length, reasoning_strength, or the tools the model uses during generation, including its built-in web and image search, which isn't charged separately. A request that returns n images is billed for n images. You're billed only for images the model successfully generates and returns: images that fail to generate, or that are removed by safety filtering before they're returned, aren't counted. Image responses still include a usage object with token counts for reference, but image generation tokens are not included in token-based pricing for other models. See the Image generation guide for the response shape.

Muse Voice Transcribe

Muse Voice Transcribe is billed by audio processed, metered in whole seconds.

Streaming and non-streaming transcription are priced the same. Zero data retention (ZDR) is priced at parity with Standard mode. Contributor mode isn't available for Muse Voice Transcribe. Audio time is billed as a separate line item from token usage, rounded down to whole seconds; other endpoints aren't billed for audio time.

Segment Anything Model 3.1

SAM 3.1 segmentation is billed per image and per video frame, not per token. The same prices apply to streaming and non-streaming requests. Each video request is billed for at least 50 frames, so a clip or tracking range shorter than 50 frames is billed as 50. The response's usage.video_frames_processed reports the frames billed for that request.

Rate limits

Rate limits apply per team, not per API key: requests, tokens, and streams from every key in a team count toward the same limits. Each model below has its own limits, so traffic to one model doesn't use up another's.

SAM 3.1 is billed per image or frame and Muse Image is priced per image, so neither has a TPM limit.

Muse Voice Transcribe is billed by audio processed, so it's limited by streams rather than requests or tokens. Concurrent streams caps how many transcription sessions you hold open at once; streams per hour caps how many you start. Realtime and file transcription share the same budget.

Token usage counts only your input and output tokens. Meta injects a small amount of steering context into every prompt (a system prompt and related scaffolding); those injected tokens are not billed and are excluded from the token counts reported in your usage and by the token-counting endpoints.

Background responses

Creating background responses (background: true) is limited to 600 submissions per minute, per team, in addition to the limits above. The cap applies only to starting new background responses: over it, new submissions return HTTP 429 Too Many Requests with a Retry-After header, and responses already running continue unaffected.

Handle rate limits

When you exceed a limit, the API rejects the request. Requests succeed again once usage drops below the limit.

Retry on 429 responses

Retry strategy: Use exponential backoff with jitter. Start with a short delay such as 500 ms, double it after each consecutive rate-limit response, and add random jitter so concurrent clients don't all retry at the same moment. If the response includes a Retry-After header, wait at least that long.

Read rate-limit headers

Every successful token-based response includes headers that report your current limit and remaining quota:

Read these headers and slow down before you hit a rate limit instead of reacting after one.

Handle Muse Voice Transcribe rejections

A realtime stream rejected for exceeding a stream limit closes with 1013. A rejected file-transcription request returns HTTP 429 Too Many Requests. Transcription responses don't include the x-ratelimit-* headers.

See also

• Models: every model line, with links to each model's versions and availability
• Subscriptions: Muse Code plans, an alternative to pay-as-you-go