AI · 7 November 2026 · 4 min read
Meta says Muse forgets less. The test is Meta’s own
Meta said on 6 November that specialised versions of its models, trained with Muse Wonder, are rolling out to Muse and Meta AI, with Active Memory and Active Focus. On its own tests, Muse shows a 2.4 times reduction in forgetting critical details, and a 67 per cent improvement on everyday tasks that hit an obstacle. There is no outside score.
Meta said on 6 November that it is rolling specialised models out to people who use Muse and Meta AI. The models were trained using Muse Wonder, which it introduced a few weeks earlier as its most capable line for personal and professional tasks. What moves across is Active Memory and Active Focus. Memory is meant to tell a lasting fact from a temporary one, and to notice when an old detail has gone stale. Focus is meant to keep going when a task hits an obstacle, try another route, or recover. Meant to is the design. It is not a week of someone’s actual use.
The numbers are Meta’s tests of Muse. It says the improved models show a 2.4 times reduction in forgetting information needed to reach a user’s goal. On an Active Memory evaluation, the current Muse model used critical context correctly on 90 per cent of tasks, against 76 per cent for the previous model. On what it calls the Personal Superintelligence Index, the current model scored 47 per cent, against 38 per cent. On tasks with an unexpected obstacle, it reports a 67 per cent improvement in finishing the job. An improvement on a company set is not an outside exam. The note does not say how many tasks were in the set, how they were chosen, or who now has the models.