Meta is moving quickly with its Muse family of AI models. Less than a month after debuting Muse Wonder and Muse Wonder Full, the company has released specialised models trained on top of Muse Wonder and designed specifically for its personal AI assistant.
Meta also introduced a new small model called Muse Ember, aimed at the high-volume, repetitive tasks that AI agents perform. The updates were highlighted in a roundup of AI news published by Hindustan Times.
Specialised Models Built on Muse Wonder
When Meta first launched Muse Wonder and Muse Wonder Full, it framed them as a significant step toward what it calls personal superintelligence. The new specialised models build on that foundation and are tuned for behaviours Meta considers central to Muse, its personal intelligence product.
Two capabilities stand out. Active Memory is meant to help the model judge how relevant earlier information remains as a conversation or task progresses. Active Focus trains the model to keep moving through a task even when it runs into obstacles along the way.
What the Benchmarks Show
Meta shared several figures comparing the new generation with the previous one. Its Personal Superintelligence Index rose from 38% to 47%, while scores on active memory evaluations climbed from 76% to 90%.
The company also reported a 67% improvement in completing tasks after hitting a problem mid-way, and a 2.4x reduction in how often the model forgets critical information. These are Meta's own measurements and have not been independently verified.
Better Multitasking and Mid-Task Adjustments
In a research post, Meta said the specialised models learned to handle multiple tasks at once and to judge when to work toward a user's goals independently versus when to collaborate in real time.
The company also said the models cope better with corrections and clarifications delivered while a task is already running, without losing track of what they were doing. For an assistant expected to carry out longer jobs, that kind of flexibility is a practical requirement.
Meta AI Gets Similar Upgrades
Muse itself is not yet available in every market, including India, so many users cannot test the new capabilities directly. Meta's general-purpose chatbot, Meta AI, is more widely available and is receiving the same set of improvements.
According to Meta, those upgrades include breaking complex problems into clear steps and asking more pointed follow-up questions to understand what a user actually wants.
Muse Ember Targets Agent Workloads
Alongside the specialised models, Meta announced Muse Ember, which it describes as its smallest model and one that delivers more intelligence per dollar than rivals in its class.
Meta CEO Mark Zuckerberg said the model was "built for token-heavy work that agents repeat all day." That includes tool calls, coding subtasks, routing, data extraction and parsing long documents. Muse Ember is available through the Meta Model API and in Muse Code.
What It Means for Developers and the AI Race
The rapid release cadence shows Meta pushing to stay competitive as the leading AI labs head into 2027. Specialised models for its assistant and a cheaper model for agent workloads point to a two-track strategy covering both consumers and developers.
For startups building AI agents, low-cost models like Muse Ember could matter more than headline benchmarks, since repetitive tasks at scale are where compute costs add up fastest. Whether Meta's reported gains hold up in real-world use will become clearer as access widens.
Source: Hindustan Times
