Overview
- Meta released two new models, Muse Wonder and the larger Muse Wonder Full, to power agent tasks that retain context and pursue multi‑step goals using features called Active Memory and Active Focus.
- Muse Wonder is available through the Meta Model API and Muse Code and is priced at $4 per million input tokens and $20 per million output tokens, while Wonder Full is restricted to Muse Code’s $50 (Power) and $200 (Maximum) subscriber tiers with no public API.
- Meta published benchmark results showing strong memory, coding and long‑horizon performance and cited third‑party scores from Artificial Analysis, but other firms including Abacus.AI have disputed the comparisons and called for full independent evaluations.
- Meta says the models include its toughest safeguards yet against chemical, biological, radiological, nuclear and cyber misuse and are trained to refuse risky requests, yet early reports flagged security bugs and enterprises continue to voice privacy and integration concerns.
- The launch follows a year of heavy investment and the September rollout of the Muse personal agent, and Meta plans a staged upgrade of Muse for subscribers while the industry watches for independent validation and broader enterprise adoption.