Key points
- Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full on October 12; Wonder is on the Meta Model API and in Muse Code, while Full is limited to Muse Code subscribers.
- Wonder API pricing is $4 per million input tokens and $20 per million output tokens, up from $1.25 and $4.25 for Muse Spark 1.3, per The Decoder.
- Artificial Analysis scores Wonder Full 56 and Wonder 52 on its Intelligence Index, versus 48 for Spark 1.3; Meta says Wonder will also power the Muse agent.
- Both models keep a one-million-token context window and accept text, image, video and audio input; Wonder Full has no published token price yet.
Why it matters for marketers: Marketers building on agents now have a third frontier-class option at mid-range pricing, but Wonder Full is gated and Meta's strongest claims come from its own benchmarks.
Meta Superintelligence Labs released two models on October 12, Muse Wonder and Muse Wonder Full, which the company positions as its first frontier-level systems. Wonder is available through the Meta Model API and Muse Code, Meta's AI coding tool. Wonder Full is limited to Muse Code subscribers and is not served through the API. Meta says Wonder will also power its Muse personal agent, starting with paying subscribers.
The headline features are Active Memory, which keeps key user information across days and weeks, and Active Focus, which helps the model stay on a task, try other approaches and ask clarifying questions. Artificial Analysis puts Wonder Full at 56 on its Intelligence Index and Wonder at 52, up from 48 for Muse Spark 1.3. The biggest gains are in agentic coding. Much of the comparison data with rival models comes from Meta's own testing, and Wired notes that benchmarks do not always reflect real-world performance.
For marketing and content teams, cost is the practical detail. The Decoder reports Wonder costs about $4.52 per task on the Intelligence Index, below Anthropic's Claude Fable 5.1 and Opus 5.5 but above GPT-6 Astra. It is also nearly three times the cost of Spark 1.3. The same analysis finds weaker factual knowledge scores than top rivals, offset by lower hallucination rates, which is worth checking for any content that must be accurate.
Watch for a Wonder Full price and API access, since subscriber-only availability limits who can test it. Also watch for independent real-world results and for how Meta handles agent safety. Wired notes Meta has not said whether Wonder-powered agents misbehaved in testing. Teams piloting agents should run their own tasks before switching models.
Sources
This brief is our own summary. Read the full reporting at the source.
- TechCrunch: Meta releases its first frontier model, Muse Wonder
- Wired: Muse Wonder Puts Meta Back at the AI Frontier
- The Decoder: Meta's new frontier model Muse Wonder Full breaks into the AI top tier
- Latent Space: [AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
- The Rundown AI: Meta's Watermelon ripens into a frontier model
- Ben's Bites: Zuck’s wonder...full
- MarkTechPost: Meta Launches Muse Wonder and Muse Wonder Full: Benchmarks, Pricing and How to Use It
