Last verified
META'S MOST CAPABLEPREVIEW1M CONTEXTAGENTIC CODINGNO LONG-CONTEXT PREMIUM

Muse Wonder API Pricing

Muse Wonder is the model Meta calls its most capable, released in preview on the Meta Model API on October 12, 2026 for complex software engineering and long-horizon agentic work. It costs $4.00/M input, $0.20/M cached input and $20.00/M output, one rate across its 1M-token window. That is several times Muse Spark 1.3, and unlike Spark there is no cheaper Contributor tier. Checked by hand against ai.developer.meta.com.

Input - per 1M tokens
$4.00/M
3.2x Muse Spark's input launch rate
Output - per 1M tokens
$20.00/M
5x the input rate launch rate
Cached input - per 1M tokens
$0.20/M
No cache-write charge published -95%
Effective - agentic blend
$2.41/M
92/8 split - 82% cache
§ 01 / TERMINAL

Run the numbers.

Live calculator on Meta's Standard rate, the only one Wonder has. Run the same workload on Muse Spark 1.3 before committing: Meta's own advice is to keep Spark for high-volume and cost-sensitive work and bring in Wonder where capability matters more than price per token.

$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
Open full calculator (all models · share URL · CSV) →
§ 02 / SCENARIOS

Real-world presets.

CODING AGENT

Codebase-scale migration

$0.306/task
120,000 in - 10,000 out~326 units/$100
LONG DOC

Reading 100-page contracts

$0.319/doc
180,000 in - 8,000 out~313 units/$100
RAG SUPPORT

Ticket triage

$0.027/ticket
8,000 in - 1,000 out~3,693 units/$100
ASSISTANT

Research planning turn

$0.056/turn
18,000 in - 2,000 out~1,788 units/$100
§ 03 / TOKENIZER

Paste text. See tokens. See cost.

What's a token →
Estimate · meta-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type

This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.

Characters 882
Words 138
Tokens (estimated) 229 tokens
Cost as input · uncached $0.00092 USD
Cost as output · uncached $0.00458 USD
Cost as cached input $0.00005 USD
§ 04 / SHELF

Up against the shelf.

All models →
Model Input /M Output /M Effective blended Context Best for
Muse Spark 1.3 $1.25 cache $0.15 $4.25 $0.66 −3.7× 1M Meta's high-volume model - same endpoints, a fraction of the rate
Muse Spark 1.2 $1.25 cache $0.15 $4.25 $0.66 −3.7× 1M Audio work on Meta's API
Claude Opus 5.5 $4.00 cache $0.20 $20.00 $2.58 +7% 1M Frontier coding at an identical list rate
Gemini 4 Argon $2.00 cache $0.10 $10.00 $1.21 −2.0× · promo 1M Google's flagship reasoning model
GPT-6.1 Sol $2.00 cache $0.10 $10.00 $1.29 −47% 1.05M OpenAI's coding workhorse
Claude Sonnet 5.5 $2.00 cache $0.10 $10.00 $1.29 −47% 1M Agent loops at half the frontier rate

Frequently asked.

What Wonder costs against Muse Spark and its frontier peers, why there is no Contributor discount, and what the preview label and Muse Wonder Full mean for API users.

Q · 01 How much does Muse Wonder cost? +
Meta lists $4.00/M input, $0.20/M cached input and $20.00/M output, Standard mode only. Under this site's 92/8 agentic blend at an 82% cache-hit rate that is $2.41/M effective. Meta states there is no long-context premium: the rate is the same whether the 1M-token window is nearly empty or nearly full.
Q · 02 How does it compare with Muse Spark? +
Wonder is the expensive end of Meta's lineup. Muse Spark 1.3 costs $1.25/M input and $4.25/M output, so Wonder's input is 3.2x and its output 4.7x Spark's. Both take the same requests on the same endpoints, so switching is a model ID change. Meta's own split: Wonder for multi-file coding changes, multi-hour agent runs and sub-agent orchestration; Spark for sub-agent tasks, classification and tight tool loops.
Q · 03 Is there a Contributor (train-on-my-data) discount? +
Not on the API. Meta's pricing table lists Wonder's Contributor mode as "Not available" and says it "isn't available for Muse Wonder on Model API". Muse Spark 1.3 and 1.2 keep their Contributor tiers. Inside Muse Code, subscribers can run Wonder in Contributor mode for more usage, but that is a subscription allowance, not a per-token price.
Q · 04 How does it compare with Claude Opus 5.5 and GPT-6.1 Sol? +
On list price, Wonder matches Claude Opus 5.5 exactly: $4.00/M input, $20.00/M output and $0.20/M cached. GPT-6.1 Sol lists at half that, $2.00/M input and $10.00/M output. Gemini 4 Argon is at half that through November 30, 2026, and Google's schedule puts it on Wonder's exact card from December 1. Unlike Anthropic, Meta publishes no Batch discount for Wonder, so there is no cheaper asynchronous route on the same model.
Q · 05 What is Muse Wonder Full, and can I call it? +
Not through the API. Muse Wonder Full is the higher-capability variant Meta announced the same day, and it runs only inside Muse Code on the Power Usage and Maximum Usage subscriptions. It has no per-token price, so it is not on this site. Meta notes that setting reasoning effort to max on muse-wonder-1.0 does not select it.
Q · 06 What does "preview" mean here? +
Meta labels the model "in preview" on its models page while it is fully priced and rate-limited on pay-as-you-go: 3,000 requests and 4,000,000 tokens per minute, the same as Muse Spark 1.3. Meta publishes no end date for the preview and no planned price change. This page will record any change once it appears on Meta's pricing page.
Q · 07 What does it cost to add web search? +
Search grounding is billed separately at $2.50 per 1,000 queries, the same as on Muse Spark, plus the input tokens the results add to your context.
Y
Reviewed by Yaroslav Vikhariev Founder · AI//COST · Prices checked by hand against ai.developer.meta.com · Last verified Oct 13, 2026
Methodology Report a correction More by Y.V.