Meta's Muse Wonder joins the AI race, and it remembers your customers

On Monday, October 12, Meta introduced two models, Muse Wonder and Muse Wonder Full, in a post from its research lab. Meta calls them its most capable models yet, built for "personal and professional agentic tasks." Wonder is available today through Meta's developer API and its Muse Code coding tool. Wonder Full, the stronger one, is only for Muse Code subscribers for now.

Meta says Wonder will soon power its Muse agent app, starting with paying subscribers. Muse is the agent Meta launched in September that can send emails, book travel and fill out forms for a person, asking for approval before it sends an email or makes a purchase, Fox Business reported.

So the model got smarter, and the assistant that shops and books for people is next in line to use it. Hold that thought.

On the Artificial Analysis Intelligence Index, an independent scorecard, Wonder Full scored 56, tied with Anthropic's Claude Sonnet 5.5 and two points behind Claude Opus 5.5 at 58, VentureBeat reported. The regular Wonder scored 52. OpenAI's GPT-6 Astra and Google's Gemini 4 Argon each scored 53.

Meta's own table is more flattering, as company tables tend to be. On a business-workflow test called AutomationBench, Wonder Full scored 45.4 against 41.6 for OpenAI's GPT-6 Astra and 41.2 for Anthropic's Claude Opus 5.5. On another coding test, Terminal-Bench 4.0, it trailed both, per VentureBeat. Meta mixes third-party and internal tests, so read it as "at the frontier," not "the winner."

It is also not cheap. Artificial Analysis lists Wonder among the most expensive models, and Wonder could come with a monthly subscription of $20 to $100, Euronews reported.

Meta built two features into Wonder. Active Memory carries what matters about a person across conversations and pulls it back up when it becomes relevant. Active Focus keeps the agent working through blockers toward a goal, and checks in before important actions, Meta wrote.

Now picture a buyer using one. An agent that remembers your customer prefers weekend appointments, hates hidden fees and already tried two competitors. When it goes looking for a dentist, a contractor or a skincare brand, it is not starting from zero. It is starting from a shortlist and a grudge.

That raises the bar for the basics. Clear pricing language, real availability, easy booking and consistent details across your website and listings are what an agent can actually read and act on.

Meta says the goal is to bring this to "the billions of people who use our products," per its post. Most of your customers already spend time in Meta's apps. An assistant with memory, living where people already scroll and message, is a very different shelf than a chatbot they have to go open.

Meta has not given dates for Instagram or Facebook, and we would not guess. What it has shown is the direction: assistants that remember, plan and buy.

For teams that build their own tools, the Wonder API costs $4 per million input tokens and $20 per million output tokens, and Muse Code plans with Wonder Full start at $50 a month, VentureBeat reported. In Artificial Analysis testing, Wonder averaged about $4.52 per task, against $5.98 for Claude Opus 5.5 and $3.26 for GPT-6 Astra.

For most small businesses, none of that is a reason to switch tools. If you already pay for ChatGPT or Claude and it works, keep it. The useful lesson is competition: four labs are trading the top spot, prices are moving, and no assistant is going to own every customer.

Meta also says paying API customers' content is not used for training, while Muse Code subscriptions default to a mode that may use submissions to improve models, per VentureBeat. If your team pastes client or customer details into any AI tool, check that setting first. Boring advice, saves careers.

And one more reason to watch this space: Meta frames all of it as a step toward "personal superintelligence," in its own words. Translated for a marketing budget, that means more assistants acting for your customers, sooner than most plans assume.

Four companies now sit within a few points of each other at the top of the AI scoreboard. The model race matters less to you than the shift underneath it: assistants are moving from answering questions to remembering people and acting for them. The businesses that win will be the clearest ones to choose.

If you want help getting your business ready for AI assistants, our founders will walk you through it. Talk to them below. See you in the next edition.