Meta's Muse Wonder lands near the frontier at $4 in and $20 out per million tokens. The stronger Full version is not in the API

Meta now has a frontier class model you can call from an automation, but read the access and pricing details first.

What happened

Meta released Muse Wonder and Muse Wonder Full, which it calls its most capable models to date, built for agentic tasks. Muse Wonder is available through the Meta Model API and Muse Code. Muse Wonder Full is only for Muse Code subscribers and, per TechCrunch, will not be served through the API.

Through the API, Muse Wonder costs $4 per million input tokens and $20 per million output tokens. The Decoder reports that its predecessor, Muse Spark 1.3, cost $1.25 and $4.25, and that Artificial Analysis measured Muse Wonder at $4.52 per task on its Intelligence Index, nearly three times the $1.60 of Muse Spark 1.3. On the same index Muse Wonder Full scores 56 and Muse Wonder 52, up from 48. Both models support a one million token context window and accept text, images, video and audio.

The weak spot is factual knowledge: on AA-Omniscience, Muse Wonder Full scores 30 and Muse Wonder 25, behind Claude Opus 5.5 at 46. Muse Wonder Full did cut its hallucination rate from 33 to 23 percent. Meta also pitches Active Memory, carrying details across conversations, and Active Focus, sticking with hard tasks and checking in before anything important. The Verge notes that Meta's comparisons with rival models come from its own internal testing.

My take

More competition at the top is good for anyone paying API bills. But for automation work, three details matter more than the headline.

• The version you can call from n8n, Make or your own code is the standard Wonder, not Full.
• Weak factual recall means you should feed it the facts, such as CRM records, product docs or the ticket history, rather than trusting what it knows.
• Judge it by cost per finished task on your own workflow. Per token price alone hides how many tokens a model burns.

Keep the model a swappable setting in your workflows, run a small test set of real client tasks through it, and switch only when the numbers say so.