TL;DR
Get audio and creator gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Meta announced Muse Wonder and Muse Wonder Full, new models it describes as its most capable for personal and professional agent tasks. Muse Wonder is available through the Meta Model API and Muse Code; Muse Wonder Full is available to Muse Code subscribers. Meta says the models are designed to retain relevant context and work through longer tasks, but independent performance evidence and broader consumer availability were not provided in the supplied report.
Meta has introduced Muse Wonder and Muse Wonder Full, two artificial intelligence models it says are built for personal and professional agent tasks, including coding, memory and work that requires sustained effort. The company says Muse Wonder is available through its Model API and Muse Code, while Muse Wonder Full is available to Muse Code subscribers; consumer experiences in Muse are planned for a later rollout.
Meta presents the models as the latest stage in a rebuilding of its AI program that began 14 months ago. The company says the work spans pretraining and reinforcement learning, and that the models were trained at its Prometheus data center. It describes both as its most capable models to date and says they achieve state-of-the-art results across selected benchmarks for agents, memory and coding. Those performance descriptions are Meta’s claims, based on evaluations it says are detailed in an accompanying report.
The company highlights two capabilities it calls Active Memory and Active Focus. Active Memory is intended to carry useful information about a person across conversations and retrieve it when relevant. Active Focus refers to continuing work toward a goal, trying different approaches when blocked, responding to new information, seeking clarification and checking with the user before important actions. Meta says it evaluated these areas with FocusBench and SWE-Marathon, among other tests.
Meta also points to evaluations in law, finance and health, and to software-development tasks that involve changes across a codebase and repeated testing. The announcement includes a demonstration in which Muse Wonder builds a playable 3D kitchen game from customizable image inputs. The report describes intended capabilities and benchmark results; it does not, in the supplied material, provide independently verified assessments of the models’ performance in everyday use.
Where Muse Wonder Will Be Used
The announcement matters because Meta is positioning the models for tasks that go beyond answering a single prompt. A system that can retain relevant context and continue work across multiple steps could support personal assistants and workplace agents, as well as coding and artifact creation. Meta says Muse Wonder will soon power experiences in its Muse product, initially for subscribers, and will also be used for artifact generation and processes requiring more intelligence.
For users and developers, availability differs by model. Muse Wonder is accessible through the API and Muse Code, while access to Muse Wonder Full is currently described as limited to Muse Code subscribers. Meta says it plans to use the models to develop specialized versions for people using Muse, but has not supplied a schedule for that broader deployment. The practical impact will depend on how reliably the models handle tools, personal information and lengthy tasks outside curated demonstrations.
As an affiliate, we earn on qualifying purchases.
Meta’s Scaling Plan for Muse
Meta frames Muse Wonder as a continuation of its announced approach to scaling AI models. The company says it introduced Muse Spark as an earlier step in a program emphasizing predictable scaling and rigorous development. It now says improvements to both pretraining and reinforcement learning underpin the Wonder series. The supplied announcement does not give a release date for Muse Spark, nor does it provide enough information to compare the models’ costs, size or availability across regions.
The report also discusses performance on familiar versus unfamiliar tasks. Meta says Muse Wonder Full performs better on unfamiliar tasks than a trend based on Muse Spark 1.3 would predict, attributing the difference to stronger underlying models. That is the company’s interpretation of its evaluation results. Meta argues that generalization matters for agents because users’ goals, tools and constraints can vary in ways developers cannot anticipate in advance.
“Muse Wonder brings together Active Memory and Active Focus to keep track of what matters to you and work diligently toward what you want to accomplish.”
— Meta, in its Muse Wonder announcement
personal AI memory assistant device
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Testing and Rollout Details
The supplied report does not establish how the models perform in independent evaluations or in routine use by customers. Meta’s benchmark claims have not been independently verified in the source material, and the report does not provide enough detail here to assess evaluation design, error rates or comparative performance across every task. It is also unclear how the models handle sensitive information retained through Active Memory, what user controls will be available, and when the planned Muse rollout will begin.
Meta says Muse Wonder Full is available to Muse Code subscribers but does not specify access by location, any usage limits or whether the model will become available through the API. The company also has not stated when the specialized models it plans to develop will reach Muse users. These details may affect how broadly people can use the models and what kinds of work they can delegate.
As an affiliate, we earn on qualifying purchases.
Planned Muse Product Rollout
Meta says Muse Wonder will soon begin powering experiences in Muse, with the rollout starting for subscribers. It also plans to use the model for artifact generation and more demanding processes, and says it intends to build specialized models for people using Muse. The announcement gives no firm dates for those steps, so the next concrete developments are likely to be additional access details, product integrations and further evaluation information from Meta.
For now, developers can access Muse Wonder through the Meta Model API and Muse Code, while Muse Code subscribers can use Muse Wonder Full, according to the company. Whether the models meet Meta’s stated goals across real-world tasks will become clearer as access expands and users and independent evaluators report results.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are Muse Wonder and Muse Wonder Full?
They are two AI models in a new series announced by Meta. The company says they are designed for personal and professional agent tasks, including memory, coding and work requiring multiple steps.
How can people access the models?
Meta says Muse Wonder is available through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. The announcement does not specify all regional or usage conditions.
When will Muse Wonder be available in the Muse app?
Meta says Muse Wonder will soon power experiences in Muse and that the rollout will start with subscribers. It did not announce a specific date.
What does Meta mean by Active Memory and Active Focus?
Meta describes Active Memory as carrying relevant personal context between conversations and retrieving it when useful. Active Focus refers to continuing a task through obstacles, adapting to new information and asking for clarification or approval when needed.
Have the models’ performance claims been independently confirmed?
The supplied announcement reports Meta’s own benchmark results and characterizes the models as state of the art. It does not include independent verification, so performance in broader real-world use remains to be assessed.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
