October 13, 2026

A fourth frontier lab is selling agent models. Don't flip the switch yet.

On Oct 12, Meta launched Muse Wonder and Muse Wonder Full for long agent tasks. Wonder is on the Meta Model API and Muse Code at $4 / $20 per million tokens. Wonder Full is Muse Code only. Test one workflow before you switch anything.

What Meta launched

On October 12, 2026, Meta introduced Muse Wonder and Muse Wonder Full, which it calls its "most capable models" for personal and professional agentic tasks. An agent here means a model that works through a multi-step job for you, across apps and files, rather than answering one question. With OpenAI, Anthropic and Google already selling models like this, Meta is now a fourth big lab in that market.

Meta pitches two capabilities. Active Memory carries important context across conversations and pulls up the details a task needs. Active Focus keeps the agent working through blockers and, in Meta's words, has it check in "before taking important actions." Those are Meta's descriptions, not results measured in a small business.

Meta launch graphic: the words Muse Wonder in blue, centered on a grid of the word WONDER repeated in pale type. — Figure: Meta's launch art for Muse Wonder. Image: Meta, Introducing Muse Wonder and Muse Wonder Full (Oct 12, 2026).

Who can use which model

01
Muse Wonder.
Available now through the Meta Model API and Muse Code. Per Meta's developer post, it costs $4 per million input tokens and $20 per million output tokens (USD), with cached input at $0.20 and a 1M-token context window. This is the model you can build a business workflow on today.
02
Muse Wonder Full.
Available only inside Muse Code, Meta's coding assistant, on the Power Usage and Maximum Usage subscription plans. There is no public API for it. VentureBeat reports Power Usage at $50 a month (20 times the Everyday plan's allowance) and Maximum Usage at $200 a month (80 times).
03
Muse, the app.
Meta says Muse Wonder "will soon power experiences in Muse," rolling out to subscribers first. It is not there yet.

Diagram titled Where Muse Wonder and Wonder Full ship. Model A, Muse Wonder: Meta Model API and Muse Code, $4 / $20 per 1M tokens, cached input $0.20, context 1M tokens, Artificial Analysis index 52. Model B, Muse Wonder Full: Muse Code only, not on the public API, Power $50 a month and Maximum $200 a month, index 56, tied with Claude Sonnet 5.5, no downloadable weights. Below: Active Memory carries context across conversations; Active Focus recovers from errors and checks in before important actions. — Figure: where each model ships, as of Oct 12, 2026. A Consultiply diagram, not a product screenshot. Availability and API prices per Meta's research and developer posts; index scores and plan prices per VentureBeat.

How good is it?

VentureBeat reports that Artificial Analysis, an independent benchmarking firm, scores Muse Wonder Full 56 on its Intelligence Index, tied with Claude Sonnet 5.5, and Muse Wonder 52. Claude Opus 5.5 scores 58. So Meta's strongest model is near the top, but it is the one you cannot reach through the API, and the API model scores lower. A general index is also not your workflow: it says nothing about how either model handles your quotes, invoices or customer emails.

What to check before you trust it

Meta says a full report with additional safety evaluations "will be available soon." Until it is out, keep Wonder away from customer records, payments and private files. VentureBeat adds two details worth knowing. Neither model has downloadable weights, so you cannot run them on your own servers. And while content sent through the standard API is not used for training, Muse Code subscriptions default to Contributor mode, which lets Meta use what you submit to improve future models. You can turn it off, but only for future submissions.

If you tried Meta's earlier agent for business apps, our Oct 6 post on Muse for Small Business covers connecting it to tools like QuickBooks and Shopify. This post is about the new models underneath, and whether to build on them.

Should you switch this week?

Decision path. Question 1: do you have one hard workflow you can test on two models this week? If no, pick that workflow first and write the brief and success score. Question 2: can you keep this off customer and confidential data until Meta's full safety report? If no, wait or use a sandbox only. Question 3: can you ship on Muse Wonder through the Meta Model API? If no, note that Full is Muse Code only and that path is not open yet. Yes to all three leads to this week's owner steps: pick one highest-friction workflow and write the brief; run the same job on Wonder and one other model you already use; name who hits the brake and keep customer data offline until the safety report. — Figure: three questions before you touch Wonder, and this week's steps. Product facts per Meta, Oct 12, 2026; the decision path is Consultiply guidance for small business owners.

01
Pick one hard workflow.
The job that already eats time or breaks often. Write the brief and what a good result looks like before you open an API key.
02
Run it on two models.
Same brief, same sample data, on Wonder and on a model you already use. Score quality, rework and where a person had to step in.
03
Name who hits the brake.
One person can pause the agent. Keep customer data out of the test until Meta's full safety report is published.

Our take

More strong models means better prices and more choice for a small business. That is good news, not a reason to rebuild your setup in launch week. Test one workflow, keep a person on the brake, and wait for the safety report.

No partnership with Meta is implied here. This is an owner's reading of Meta's announcement and the coverage around it.