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 has introduced Muse Wonder and Muse Wonder Full, models it says are designed for personal and professional agent tasks, coding and memory. Muse Wonder is available through the Meta Model API and Muse Code, while Wonder Full is available to Muse Code subscribers. Meta has not supplied independent verification of its performance claims in the provided report.
Meta has introduced Muse Wonder and Muse Wonder Full, two AI models the company says are built for personal and professional agent tasks, including coding and work that requires retaining context. Meta says Muse Wonder is available through its Meta Model API and Muse Code, while Wonder Full is currently available to Muse Code subscribers.
In its announcement, Meta describes the models as its most capable to date and says they achieved state-of-the-art results across selected benchmarks for agents, memory and coding. Those are company-reported performance claims; the supplied report does not include independent testing or enough benchmark figures to compare results directly. Meta says its evaluation suite includes legal, finance and health tasks, as well as tests of sustained software development.
The company presents two design ideas as central to the models. Active Memory is Meta’s term for carrying useful information across conversations and retrieving it when relevant. Active Focus describes continuing to work through a challenging task, trying other approaches when blocked, adapting to new information, and asking questions or checking in before important actions.
Meta says Muse Wonder will soon power experiences in its Muse product, with an initial rollout to subscribers. It also plans to use the model for artifact generation and processes it describes as requiring greater intelligence, and says it may later develop specialized models from Muse Wonder for broader use. The announcement does not give a schedule for those plans.
A Move Toward Persistent AI Agents
The announcement matters because Meta is positioning Muse Wonder as more than a model for single prompts: it is intended to carry context between conversations and sustain work over multiple steps. If the described functions work reliably, they could make assistants more useful for projects that involve changing requirements, files, tools or follow-up decisions.
That design also raises practical questions about how an agent uses personal information and when it acts without further input. Meta says the model asks for clarification and checks in before important actions, but the announcement does not specify the controls users will have, what kinds of actions will require approval, or how retained context will be managed. Those details will matter as the model reaches consumer-facing products.
For developers, availability through the Meta Model API and Muse Code offers a route to try the model in software and agent workflows. However, the report does not provide pricing, usage limits or independent comparisons. Its benchmark claims indicate Meta’s intended position in a competitive market, but do not by themselves establish how the models will perform in customers’ particular tasks.
As an affiliate, we earn on qualifying purchases.
Meta’s Scaling and Product Plans
Meta frames Muse Wonder as a continuation of work it says began with a ground-up overhaul of its AI efforts 14 months before the announcement. The company says the release builds on its earlier Muse Spark work and advances its approach to scaling pretraining and reinforcement learning. It also says the new models were trained at its Prometheus data center, which Meta describes as gigawatt-scale.
The report organizes its evaluation discussion around domain knowledge, memory, sustained focus and coding. It names Harvey’s Legal Agent Benchmark as one legal evaluation and refers to FocusBench and SWE-Marathon for sustained task performance. Meta also discusses performance on unfamiliar tasks, arguing that stronger base models may help post-training improvements carry over beyond tasks targeted during training. This remains the company’s interpretation of its own evaluation results.
Meta’s stated longer-term aim is to bring more capable assistants to people using its products. In the near term, its announcement distinguishes access to the two models: Muse Wonder is available through the API and Muse Code, while Wonder Full is limited to Muse Code subscribers. Broader Muse availability is described as upcoming, not as a completed rollout.
“Active Memory”
— Meta
Independent Results and User Controls
The announcement does not provide independent confirmation of Meta’s claims that the models deliver state-of-the-art performance. It references a broad evaluation suite and names several benchmarks, but the source material supplied here contains no numerical scores, testing methodology in full, or third-party comparisons. The models’ performance in everyday use, including on unfamiliar tasks, remains to be established beyond Meta’s reported evaluations.
Other details are also not specified: pricing and API limits, the timing of the Muse rollout, the regions or user groups included, and how Active Memory stores or deletes information. The report says the model checks in before important actions, but does not define which actions count as important or how that safeguard works across apps. These points may become clearer through product documentation and user access.
Muse Rollout and Model Access
Meta says Muse Wonder will soon power experiences in Muse, beginning with subscribers, but it gives no launch date. The next concrete development to watch is when the company opens that access and publishes further details about supported features, user controls and availability. Until then, the stated Muse rollout remains a plan rather than a released feature.
Developers and Muse Code subscribers can assess the models through the access Meta says is available now. Further information from Meta about benchmark methods, pricing, data handling and real-world performance would help users judge whether the claimed memory and sustained-task capabilities translate into dependable results. Meta also says it intends to develop specialized models based on Muse Wonder, without announcing a timeline or naming those models.
Key Questions
What are Muse Wonder and Muse Wonder Full?
They are two AI models announced by Meta. The company says they are designed for personal and professional agent tasks, coding, memory and sustained work.
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.
When will Muse Wonder arrive in the Muse product?
Meta says it will roll out soon, starting with subscribers, but the announcement does not give a specific date.
Are Meta’s performance claims independently verified?
The source describes Meta’s own evaluations and benchmark claims. It does not provide independent verification or enough numerical results to confirm those claims separately.
What does Active Memory mean?
Meta uses the term for retaining useful context across conversations and retrieving details when they become relevant to a task. The announcement does not specify the data-retention controls or deletion process.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
