Meta announced Muse Wonder and Muse Wonder Full on October 12, 2026, bringing both models to Muse Code subscribers and offering Wonder through the Meta Model API. The launch introduces new options for coding and complex tasks, with different access routes for developers and users of Meta’s coding product. (about.fb.com)

The separate rollout into the Muse agent is still planned. Meta says Wonder will power experiences there beginning with subscribers, but has not supplied a firm date. Launch-day access through Muse Code should therefore not be confused with availability in the broader Muse agent. (theverge.com)

What is available now

The practical distinction is between access through a coding product and access for building applications. Wonder is the model Meta identifies for both Muse Code and its API; Wonder Full’s announced route is a Muse Code subscription. The launch announcement itself does not identify which subscription tiers qualify. (about.fb.com)

For prospective buyers, that makes the access route a separate question from the model’s capabilities. A team evaluating a model inside Muse Code is not necessarily evaluating an option it can deploy through the API. The two model names should not be treated as interchangeable purchasing choices. (dev.meta.ai)

Meta’s current API pricing documentation lists Wonder at $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. These are usage-based API charges, not the price of a Muse Code subscription, and should be considered separately when estimating the cost of a proposed deployment. (dev.meta.ai)

What Meta says the models can do

Meta describes the Wonder family as supporting software work that involves tracking requirements, changing code across a project, and repeatedly testing and revising the result. Those descriptions are company capability claims, rather than findings from hands-on testing conducted for this article. (research.meta.ai)

The company groups two other capabilities under the names Active Memory and Active Focus. Active Memory is its term for retaining important context across conversations and retrieving details relevant to the current task. Active Focus describes persistence on difficult work: trying alternatives, recovering from errors, seeking clarification and checking before important actions. (research.meta.ai)

Meta also positions the models for work involving law, finance and health, citing examples such as understanding a lease or navigating an insurance claim. These examples describe intended uses; they do not establish that the models’ outputs are reliable enough to replace professional judgment. (about.fb.com)

Meta calls the models its most capable to date and claims state-of-the-art performance on personal and professional agent tasks. This article does not independently establish competitive superiority or reproduce benchmark rankings. Availability is a concrete launch detail; claims about which model performs best require a separate evaluation of evidence. (research.meta.ai)

A launch within a broader business push

The release follows Meta’s September 28 announcement of Meta Enterprise Platform. That initiative aims to bring its models, agents and infrastructure to businesses and developers, with Muse Code among the products named in its initial focus. The Wonder launch therefore arrives alongside a broader effort to turn Meta’s AI technology into business products. (about.fb.com)

That context helps explain why distribution matters alongside model performance. Enterprise buyers assessing Meta’s expanding lineup need to distinguish a model offered for application development from a model accessed within a packaged tool. The company’s platform announcement sets out a business direction, not evidence that any particular deployment will meet a customer’s requirements. (about.fb.com)

Long-running coding work is also a focus for competitors. In its April 16, 2026, announcement of Claude Opus 4.7, Anthropic emphasized complex software engineering, sustained execution and checking outputs before reporting back. Those were Anthropic’s own descriptions, not a direct comparison with Wonder. (anthropic.com)

The overlap illustrates a common product ambition: completing substantial assignments rather than simply generating an answer. It does not establish equivalence between the companies’ models. For buyers, similar positioning makes task-specific evaluation more useful than treating launch language as a performance ranking. (anthropic.com)

What remains ahead

Beyond the subscriber-first Muse rollout, Meta plans to use Wonder to develop specialized models for everyone using Muse. That commitment is not an announcement of universal access to either Wonder or Wonder Full, and the company has not provided a timetable for those specialized models. (theverge.com)

Meta says it trained the Wonder family to avoid disclosing private information without permission and to reject assistance with harmful acts. Its announcement also says a full report with additional safety evaluations is forthcoming. Those safeguards remain attributed company claims, with further published evidence still promised at launch. (about.fb.com)