Meta launches Muse Wonder API, reserves Wonder Full for Muse Code subscribers
The AMW Read
Meta's reported first frontier-model release meaningfully updates its foundation-model position, with API pricing and subscription-only access shaping segment competition without independently proving capability leadership.
Meta launches Muse Wonder API, reserves Wonder Full for Muse Code subscribers
Meta's Superintelligence Lab released Muse Wonder and Muse Wonder Full on October 12, positioning them as its first frontier models after the lightweight Muse Spark. Wonder costs $4 per million input tokens and $20 per million output tokens through the Meta Model API. Wonder Full is available only to Muse Code subscribers. Both models will power the Muse agent. Meta reports benchmark results comparable to Fable 5.1 and GPT-6 Astra, alongside stronger performance on complex tasks such as coding; these are company-reported results.
The release expands Meta's role in the foundation-model market from low-latency assistance toward more demanding workloads. Its distribution separates developer access through the API from access to the larger model through a subscription product, making model availability part of the competitive proposition. Meta also emphasizes "active memory," or retaining and synthesizing user information over time. Its reported MemoryBench lead adds a specific test of persistent-assistant capability, although the supplied report does not establish how that performance translates into reliable use over months.
For builders, the concrete decision is whether Wonder's performance justifies its token costs in their own workloads. Teams evaluating coding or persistent-assistant applications should test task completion and recall with representative inputs rather than treat Meta's benchmark comparisons as proof of production advantage. Access is a separate constraint: developers can evaluate Wonder through the API, but cannot assume Wonder Full is available for the same integration.
