From $1.60 To $4.52 For The Same Index Task
Meta introduced Muse Wonder and Muse Wonder Full on 12 October and calls them its most capable models so far. Artificial Analysis, which tests models on a fixed set of tasks, puts Muse Wonder at $4.52 per task on its Intelligence Index, against $1.60 for Muse Spark 1.3. That is about 2.8 times more.
| Model | Cost per task on the index |
|---|---|
| GPT-6.1 Sol | $0.72 |
| Muse Spark 1.3 | $1.60 |
| Gemini 4 Argon | $1.99 |
| GPT-6 Astra | $3.26 |
| Muse Wonder | $4.52 |
| Claude Opus 5.5 | $5.98 |
| Claude Fable 5.1 | $7.63 |
The figures are Artificial Analysis's, as reported by The Decoder on 12 October. Muse Wonder now sits above GPT-6 Astra and Gemini 4 Argon and below both Claude models, where Muse Spark 1.3 sat near the cheap end.
Token Prices Rose Three To Five Times
Through the Meta Model API, Muse Wonder costs $4 per million input tokens and $20 per million output tokens, TechCrunch reports. Muse Spark 1.3 cost $1.25 and $4.25.
| Item | Muse Spark 1.3 | Muse Wonder |
|---|---|---|
| Input per million tokens | $1.25 | $4 |
| Output per million tokens | $4.25 | $20 |
| Cost per index task | $1.60 | $4.52 |
Input rose about 3.2 times and output about 4.7 times. Muse Wonder also reads 28 percent more input tokens per task, and input makes up 76 percent of the cost, according to The Decoder, while it writes about 10 percent fewer output tokens.
The Stronger Model Has No Price At All
Muse Wonder Full scores 56 on the index, fourth among the models Artificial Analysis lists, ahead of Muse Wonder at 52 and Muse Spark 1.3 at 48. Meta offers it only to Muse Code subscribers, and Artificial Analysis records no publicly available API. Without an API there is no per-token price to compare and nothing to put in a procurement contract.
Artificial Analysis also lists the model as proprietary, with weights not publicly available. Meta's launch posts state no pricing and no regional availability, so a buyer in the EU or UK should get both confirmed in writing before planning around it.
Where The Rank Overstates The Value
The headline score hides weak spots. On AA-Omniscience, a factual knowledge test, Muse Wonder Full scores 30 and Muse Wonder 25, against 46 for Claude Opus 5.5, 43 for GPT-6 Astra and 42 for Gemini 4 Argon. Finance and legal performance stayed at the predecessor's level or fell by up to two points, The Decoder reports.
There is a gain on the other side: Muse Wonder Full cut its hallucination rate from 33 to 23 percent. Meta's own benchmark results come from its model card, so they are self-reported.
What To Do Before You Switch
Run a sample of your real tasks and work out the cost per finished task, not per token, because Muse Wonder reads more input per task than its predecessor. A rank on a leaderboard is not the price on your invoice.
If your vendor file needs a safety report, Meta says a full report with additional safety evaluations will be available soon, so that file has a gap until it lands. Ask Meta in writing for EU and UK availability and data terms, and do not plan around Muse Wonder Full until it has an API.
Read next: Gemini 4 Argon Is Priced Before It Can Be Bought | Sonnet 5.5 Keeps Its Price And Cuts The Bill
