Meta has officially entered the race for high end artificial intelligence with the debut of Muse Wonder and Muse Wonder Full. Developed by the company’s Superintelligence Lab, these represent Meta’s first true foray into frontier models, moving away from the lightweight efficiency of previous releases like Muse Spark. These newer, larger models are designed to tackle far more complex workloads, particularly in areas like advanced coding, where they boast significantly higher performance benchmarks.
Internal testing suggests that Muse Wonder can hold its own against industry giants such as GPT-6 Astra and Fable 5.1. However, this increase in capability comes with a corresponding price tag for developers using the Meta Model API, with costs set at four dollars per million input tokens and twenty dollars per million output tokens. While the standard Wonder model is widely accessible via API, the more powerful Wonder Full version remains exclusive to Muse Code subscribers.
A standout feature of this latest release is what Meta describes as active memory. By focusing on the MemoryBench benchmark, researchers have prioritized a system that allows the AI to retain and synthesize user information over extended periods rather than just within a single session. Research scientist Kavya Kopparapu noted that significant effort went into ensuring the model remembers details across days and months, pushing Meta closer to its goal of creating truly personalized superintelligence agents.
This rollout marks a pivotal shift for Meta following a restructuring of its AI division led by Alexandr Wang last August. After spending several months releasing smaller tools and agents, Meta is finally challenging OpenAI and Anthropic on their home turf with state-of-the-art scale. The heavy lifting behind these models took place at Prometheus, Meta’s massive new data center in central Ohio that became operational earlier this year.