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 introduced Muse Wonder and Muse Wonder Full, describing them as its most capable models 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 will later support features in Muse, but independent performance results and the timing of that rollout were not provided in the supplied report.
Meta has introduced Muse Wonder and Muse Wonder Full, two AI models it says are built for personal and professional agent tasks, including coding and work that requires retaining context across conversations. Muse Wonder is available through the Meta Model API and Muse Code, while Muse Wonder Full is currently available to Muse Code subscribers; Meta says Muse Wonder will later power features in its Muse product.
Meta describes the models as its most capable to date and says they were developed through expanded work on pretraining and reinforcement learning. The company says they were trained at its Prometheus data center. Its announcement frames the release as part of a broader effort to build AI agents that can help with tasks at work and in everyday life.
The company highlights two capabilities it calls Active Memory and Active Focus. Active Memory is Meta’s term for carrying relevant information across conversations and retrieving it for later tasks. Active Focus describes continuing work toward a goal, trying other approaches when blocked, adapting to new information and asking for clarification when needed. Meta also says the models check in before taking important actions.
Meta reports strong results across evaluations for domain expertise, memory, focus and coding. It names Harvey’s Legal Agent Benchmark, FocusBench and SWE-Marathon among the tests discussed. The announcement describes these evaluations and comparisons with other models, but the supplied material does not include the full numerical results or independent assessments. Meta also presents demonstrations, including a playable 3D kitchen game generated from image inputs and an agent working across a browser and apps.
Where Meta Plans to Use Wonder
The release matters because Meta is presenting Muse Wonder not just as a chatbot model, but as a basis for agents that can sustain multi-step work. Such systems may need to remember a user’s requirements, use tools, respond to setbacks and avoid acting without needed clarification. Meta’s descriptions of Active Memory and Active Focus indicate the types of behavior the company is targeting.
For users, the immediate impact is limited to the access routes Meta identifies: Muse Code and the Meta Model API, with the Full model restricted to Muse Code subscribers. The wider product impact depends on the planned Muse integration. Meta says the model will also support artifact generation and processes requiring more intelligence, and that it intends to develop specialized models from it for Muse users. It has not specified when those features will arrive or which users will receive them first beyond saying the rollout will begin with subscribers.
The announcement’s performance claims could influence developers deciding which models to test for coding or agent workflows. But Meta’s benchmark descriptions are company-reported evidence, not by themselves confirmation that the systems will perform reliably across real-world uses. Results, costs, access limits and performance on tasks outside the reported evaluations will shape the practical value.
As an affiliate, we earn on qualifying purchases.
Meta’s Scaling Effort Behind Muse
Meta says it began a ground-up overhaul of its AI efforts 14 months before the announcement. It describes Muse Wonder as a continuation of its scaling work following Muse Spark, which the company says set out an approach to predictable, rigorous scaling. The announcement says the new models build on that work through changes to pretraining and reinforcement learning.
Meta argues that agent systems need to generalize beyond tasks directly emphasized during training, since users may combine goals, tools and constraints in ways developers cannot anticipate. In its discussion, the company compares Muse Wonder Full with Muse Spark 1.3 across reasoning-effort settings and says Wonder Full performs better on unfamiliar tasks than its in-distribution results would predict. This is Meta’s interpretation of its evaluation findings; the supplied report does not provide enough numerical detail to independently assess the comparison.
““Muse Wonder and Muse Wonder Full” are described as “our most capable models” for personal and professional agentic tasks.”
— Meta, in its Muse Wonder announcement
As an affiliate, we earn on qualifying purchases.
Evidence and Rollout Still Pending
The supplied announcement does not provide the full benchmark tables, scores or evaluation methodology, so readers cannot assess the reported results in detail from this material alone. It also does not include independent tests, information about model pricing, usage limits or a full account of the tools and permissions available to agents.
The timing of the planned Muse integration is unclear. Meta says the rollout will start with subscribers but does not give a date, explain which Muse features will use Wonder first, or specify when access might expand. The announcement also does not establish how the models handle errors, sensitive information or consequential actions in deployment; those questions require product details and testing beyond the claims in the release.
As an affiliate, we earn on qualifying purchases.
Access and Muse Integration
Developers can access Muse Wonder through the Meta Model API and Muse Code, according to Meta. Muse Code subscribers can access Muse Wonder Full. Meta says Muse Wonder will soon power experiences in Muse, starting with subscribers, and may support artifact generation and other processes that require more intelligence.
The next concrete developments to watch are Meta’s product rollout, publication of complete evaluation results and disclosure of practical access terms. Until then, the announcement establishes what Meta is offering now and what it plans to build toward, but not the date or scope of broader availability.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are Muse Wonder and Muse Wonder Full?
They are two AI models in Meta’s Muse Wonder series. Meta says they are designed for personal and professional agent tasks, including coding, memory and sustained work toward goals.
How can people access Muse Wonder?
Meta says Muse Wonder is available through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers.
When will Muse Wonder be available in Muse?
Meta says it will soon power experiences in Muse, with the rollout starting with subscribers. The announcement gives no specific date.
Has Meta’s performance claim been independently verified?
The supplied announcement reports Meta’s own benchmark findings and comparisons. It does not provide independent evaluation results, and the full scores and methods are not included in the material provided.
Source: hn
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
