Meta's Muse Wonder: Higher Benchmark Scores Come With Tripled Costs

Meta is pushing deeper into the frontier model race with the release of Muse Wonder and Muse Wonder Full. According to TechCrunch and The Decoder, these models mark a massive leap from previous lightweight systems, delivering top-tier performance in coding and autonomous problem-solving.

What changes in your daily workflow?

The new models introduce features like "Active Memory" for better recall across long periods and "Active Focus" to handle complex multi-step tasks without quitting. However, as The Decoder notes, factual knowledge remains a weak spot, with Muse Wonder Full scoring only 30 points on knowledge tests compared to over 40 for competing systems from OpenAI and Anthropic.

How to access and run the new models

Access is tightly tied to specific subscriptions and developer tools:

• Muse Wonder API: Available for developers via the Meta Model API, priced at $4 per million input tokens and $20 per million output tokens.
• Muse App: Powering the personal assistant app for paying subscribers.
• Muse Wonder Full: Restricted exclusively to users subscribed to the Muse Code engineering platform.

Is it worth the higher price tag?

According to Artificial Analysis data cited by The Decoder, running a task with Muse Wonder costs roughly $4.52—nearly three times the $1.60 cost of its predecessor, Muse Spark 1.3. If your work relies heavily on coding and logical proofs, the performance jump is substantial. If you need broad factual recall, other tools still hold the advantage.