Markdown edition

# Meta releases Muse Wonder and Wonder Full for AI agents

> Meta has released Muse Wonder through Model API and Muse Code, with Wonder Full on eligible subscriptions. Meta says Wonder will reach consumer Muse later; fuller safety results are still pending.

By BIG CHANGE Editorial

Published: 2026-10-12T22:13:11.530Z
Updated: 2026-10-12T22:13:11.530Z
Canonical: https://bigchange.ai/blog/meta-muse-wonder-agent-model-access

![Conceptual charcoal illustration of a physical wedding seating plan with round table models and blank place cards, beside a set banquet table.](https://bigchange.ai/api/media/file/meta-muse-wonder-seating-plan-hero-v3.png)
AI-generated conceptual illustration by BIG CHANGE, based on a task described by Meta AI Research; not an actual Muse Wonder output.

Meta introduced Muse Wonder and Muse Wonder Full on October 12, 2026, extending its Muse line from the earlier Spark model to models it presents as capable of longer agent tasks and demanding coding work. Developers can use Muse Wonder through the Meta Model API or Muse Code. Muse Wonder Full is available through eligible Muse Code subscriptions. Meta says the models will power its consumer Muse agent later; that rollout is not part of the October 12 developer release.

## The big change

- **What changed:** Meta now offers Muse Wonder through Model API and Muse Code for developers building agents that work across multiple steps. Meta lists Muse Wonder Full for Power Usage and Maximum Usage Muse Code subscriptions, while the consumer Muse product has yet to receive Wonder.
- **Why it matters:** The access split determines who can try the model in their own tools at launch. Meta describes memory across conversations and sustained work through obstacles, capabilities that matter when an agent handles a long task rather than a single prompt. Their reliability in ordinary use remains to be shown.
- **What to watch:** Meta's promised safety report will provide more detail about its evaluations. Independent tests, when available, and the announced consumer rollout will give readers more grounds to judge how Wonder performs beyond Meta's launch materials.

## Who can use the new models

[Meta's launch announcement](https://about.fb.com/news/2026/10/introducing-muse-wonder-models/) names two models. Muse Wonder is available through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. The [developer model page](https://dev.meta.ai/models/muse-wonder) identifies the API model as `muse-wonder-1.0`, lists a one-million-token context window, and prices it at $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. It lists `muse-wonder-full-1.0` for the Power Usage and Maximum Usage Muse Code subscription plans, without an API token price for Full.

The distinction matters for readers looking for the model in Meta's consumer agent. Muse already exists; BIG CHANGE has [covered its small-business connections and approval controls](https://bigchange.ai/blog/meta-muse-small-business-connectors-approvals). Meta says Wonder will power experiences in Muse **soon**, starting with subscribers. That is a future product change, not evidence that people using Muse at launch are already talking to Wonder. Meta's [API overview](https://dev.meta.ai/docs/overview) describes Muse Wonder access as a preview, so availability and terms warrant a fresh check before adopting it.

Meta has not announced Wonder as an open-weight release. [Axios reports](https://www.axios.com/2026/10/12/meta-new-ai-models-anthropic-openai) that Meta has not said whether it will release weights for the Wonder family. The separately released Muse Glimmer has open weights; its status should not be carried over to Wonder.

## What Meta means by memory and focus

Meta's [research post](https://research.meta.ai/blog/muse-wonder) describes **Active Memory** as carrying important context across conversations and retrieving details relevant to the current task. It describes **Active Focus** as continuing a difficult task by trying another approach, working through a blocker, adapting to new information, asking for clarification and checking before important actions. Those are descriptions of intended model behavior, not a published guarantee that an agent will remember every relevant fact or finish every assignment.

The mechanisms are relevant to Meta's pitch for personal and professional agents: a useful agent needs to retain the right context while pursuing a task that may take several steps. Meta's [research post](https://research.meta.ai/blog/muse-wonder) depicts Muse Wonder preparing a custom wedding seating chart from complex matching requirements; BIG CHANGE has not reproduced the task or verified general reliability from that company demo. Meta points to evaluations in memory, coding and extended execution, including FocusBench and SWE-Marathon. Its research post and model page present comparative results, but the launch-day numbers and conclusions are largely company reported. [TechCrunch's account](https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/) also attributes the comparisons to Meta's testing. BIG CHANGE has not run the models or reproduced those evaluations.

## Performance and safety remain open questions

Meta calls Muse Wonder and Wonder Full its most capable models and says they achieve leading results on personal and professional agent tasks. The evidence available here supports the narrower statement that Meta published those results, with its chosen tasks, settings and comparisons. It does not establish how reliably the models work across other people's tools, data and approval steps. Independent reporting confirms the launch and access terms; it does not independently validate Meta's benchmark claims.

Safety is especially consequential for an agent that may work with messages, calendars or documents. Meta says it trained the Wonder family to handle private information carefully, resist harmful requests and support defensive cyber work. Both the [launch announcement](https://about.fb.com/news/2026/10/introducing-muse-wonder-models/) and [research post](https://research.meta.ai/blog/muse-wonder) say a full report with additional safety evaluations is coming. Until that report is available, readers cannot inspect the complete evaluation behind the company's safety characterization. Meta's stated intention that the model check in before important actions should be assessed in the product and task settings where those actions occur.

For now, the concrete release is developer access to Muse Wonder and subscription access to Muse Wonder Full. Its consumer deployment, broader independent performance evidence and complete safety evaluation are separate milestones still to come.

### Sources & further reading

- [Meta Newsroom: Introducing Muse Wonder and Muse Wonder Full](https://about.fb.com/news/2026/10/introducing-muse-wonder-models/) (October 12, 2026). Meta's announcement establishes the two-model release, current access split, announced Muse rollout and its stated safety approach. Product and performance descriptions here are Meta's claims.
- [Meta AI Research: Muse Wonder](https://research.meta.ai/blog/muse-wonder) (October 12, 2026). Explains Meta's definitions of Active Memory and Active Focus and presents its evaluation discussion. Its benchmark results are company reported; it says a fuller safety evaluation report will follow.
- [Meta developer model page](https://dev.meta.ai/models/muse-wonder) (checked October 12, 2026). Lists model identifiers, context window, API token prices and Muse Code access. Pricing and availability can change; check the page before using the service.
- [Meta Model API overview](https://dev.meta.ai/docs/overview) (checked October 12, 2026). Describes Muse Wonder as a preview and provides the developer access path. It does not establish that every account or region has access.
- [Axios: Meta unveils new frontier-class AI models](https://www.axios.com/2026/10/12/meta-new-ai-models-anthropic-openai) (October 12, 2026). Independent reporting on the launch and Meta's competitive positioning; clarifies that the open-weight Muse Glimmer is a separate model. It does not verify Wonder's benchmark results.
- [TechCrunch: Meta releases its first frontier model](https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/) (October 12, 2026). Independent account of the release and access terms, while attributing performance comparisons to Meta's testing. BIG CHANGE conducted no hands-on test.

## Sources

- [Meta Newsroom: Introducing Muse Wonder and Muse Wonder Full](https://about.fb.com/news/2026/10/introducing-muse-wonder-models/) — Meta's October 12, 2026 announcement establishes the two-model release, developer and Muse Code access, future consumer Muse rollout, and the company's stated safety approach. Performance and safety descriptions are Meta claims.
- [Meta AI Research: Muse Wonder](https://research.meta.ai/blog/muse-wonder) — Meta's October 12, 2026 research post defines Active Memory and Active Focus and presents the company's evaluation discussion. Benchmark results are company reported; Meta says a fuller safety evaluation report will follow.
- [Meta developer model page](https://dev.meta.ai/models/muse-wonder) — Official product page lists model identifiers, context window, Model API token prices and Muse Code subscription access. Pricing and availability can change; recheck before use.
- [Meta Model API overview](https://dev.meta.ai/docs/overview) — Official API documentation describes Muse Wonder as a preview and outlines developer access. It does not establish that every account or region has access.
- [Axios: Meta unveils new frontier-class AI models](https://www.axios.com/2026/10/12/meta-new-ai-models-anthropic-openai) — Independent launch context and clarification that open-weight Muse Glimmer is separate from Wonder. It does not independently validate Wonder's benchmark results.
- [TechCrunch: Meta releases its first frontier model](https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/) — Independent reporting on the release and access terms, while attributing comparisons to Meta testing. BIG CHANGE did not run a hands-on test.