Meta makes Muse better at remembering and more resilient: 90% accuracy on key memories, 67% improvement on blocked tasks.

Beating AI News Flash: Meta announced that personal agent Muse and Meta AI have begun using a new batch of specialized models. These models were distilled from the flagship Muse Wonder released in October, gaining two core capabilities: Active Memory and Active Focus.

Active Memory enables the AI to judge which information is still valid. The model can distinguish between long-term preferences, temporary requests, and expired information, and proactively retrieve relevant content when needed for subsequent tasks. According to Meta's internal testing, Muse correctly used key context in 90% of tasks, compared to only 76% for the previous version. Converted to error rates, this means a drop from 24% to 10% — a 2.4x reduction.

Active Focus enables the AI to keep working even when it encounters obstacles. When tools report errors, task conditions change, or users modify requirements mid-task, the model will try alternative approaches, adjust its execution plan, and confirm with the user when necessary. Meta claims that in everyday tasks involving unexpected obstacles, Muse's task completion performance improved by 67% over the previous version. It can also adjust its approach when task conditions change, judging when to push forward and when to involve the user.

After this upgrade, Muse's score on Meta's self-built "Personal Superintelligence Index" also rose from 38% to 47%.