meta has introduced two new ai models – muse wonder and muse wonder full. the company positions them as its most powerful solutions for personal and professional ai agents.
muse wonder benchmark results
according to meta's benchmark results:
- meta's new models lead in 5 out of 9 benchmarks in the personal and professional agents category, while remaining competitive in the others;
- they also stand out in context retention, supported by the newly introduced active memory feature;
- the muse models score more than 10 points below claude opus 5.5 in from-scratch coding;
- they perform better in long-horizon coding and ultra-long-horizon engineering, supported by meta's new active focus technology.
what muse wonder means for ai agents
muse wonder and muse wonder full could mark a significant development in autonomous ai agents, workflow automation and long-horizon software engineering. the models outperform competitors from anthropic and openai in several benchmarks, although their performance varies across tasks.
the launch marks a notable step in meta's ai ambitions, with its real-world impact depending on how well these capabilities translate into practical applications.
sources
- muse wonder announcement, meta ai research blog – oct 12, 2026
- muse wonder model card, meta – benchmark methodology; compared with gpt-6 astra, claude fable 5.1 and claude opus 5.5
