Meta officially released its advanced Muse Wonder and Muse Wonder Full AI models, targeting both consumer personal assistants and professional agent workflows. The models feature “Active Memory” for cross-dialogue continuity and “Active Focus” to handle multi-step, long-horizon tasks without stalling. Trained on massive cluster data centers, the new lineup scores competitively against leading frontier models like Claude Fable and GPT variants on various agent benchmarks. Meta is simultaneously pushing commercialization through its Muse Code subscriptions and API access, stepping up direct competition in the paid enterprise and developer markets. Early user testing reveals a mixed reception, highlighting immense potential in automated coding alongside some real-world execution hiccups and concerns over token consumption.

Full report : Meta debuts its powerful new “Watermelon” models, aiming to rival top industry competitors with breakthrough active memory and long-horizon task execution.