The models are intended for agent tasks including document work, tool use, planning, research, and software development.
According to the report, Meta could not rule out that Muse Wonder Full would meet its high-risk threshold for chemical and biological misuse if safeguards were absent. Meta says it applied safety post-training and system-level classifiers, and assessed residual risk across the evaluated areas as moderate or lower with those protections in place.
The report also says Meta acknowledges inaccurate or biased responses, multi-step reasoning errors, and differences in performance by language and deployment setup. Meta recommends human confirmation before irreversible actions. Muse Wonder is available through the Meta Model API and Muse Code; Muse Wonder Full is available to Muse Code subscribers.
The source does not provide an independent validation of Meta’s thresholds, safeguards, or residual-risk ratings. It says Meta plans a fuller report with additional safety evaluations. Source: https://www.helpnetsecurity.com/2026/10/13/meta-muse-wonder-full-models