Short answer: On 12 October 2026 Meta introduced Muse Wonder and Muse Wonder Full, its most capable models so far, built for personal and professional agent tasks. Muse Wonder is available now via the Meta Model API and Muse Code; Wonder Full is for Muse Code subscribers; both will power the Muse personal agent, starting with subscribers.

By Agent Playbox Editorial · Published

Meta Muse Wonder: Meta's new frontier models for agents

Key facts

What did Meta announce on 12 October 2026?

Meta introduced the Muse Wonder series, two models it describes as its “most capable models to date”, aimed at what the company calls personal and professional agentic work (Meta newsroom, 12 Oct 2026). Mark Zuckerberg shared the launch on X the same day, writing that Muse Wonder is “our new family of frontier models with state of the art performance for personal and professional agentic work” (X post).

The family has two members. Muse Wonder is the version developers can call today through the Meta Model API and use in Muse Code, Meta’s coding assistant. Muse Wonder Full is a larger, higher-capability variant that is available to Muse Code subscribers. Meta says Muse Wonder will “soon power experiences in Muse”, its consumer personal agent, rolling out first to subscribers.

What can Muse Wonder actually do?

Meta frames the models around everyday, high-stakes paperwork as much as coding: its examples include navigating an insurance claim, making sense of a lease, understanding someone’s finances and building a working app. The company says the models extend that ability into expert domains such as law, finance and health.

Active Memory and Active Focus

Two named capabilities are central to the launch. Active Memory is meant to carry what matters to a user across conversations and bring back relevant details when a task needs them. Active Focus is designed to keep an agent working toward a goal when a task gets hard: trying other approaches, recovering from errors, asking for clarification and checking in before anything important.

Inputs and context

According to VentureBeat and The Decoder, both models accept text, images, audio and video, return text, and support a context window of roughly one million tokens, enough to hold long contracts, codebases or months of notes in a single session.

How does Muse Wonder compare with GPT-6 Astra and Claude?

Independent evaluator Artificial Analysis scored Wonder Full at 56 and Wonder at 52 on its Intelligence Index, as reported by VentureBeat. That places Wonder Full level with Claude Sonnet 5.5, ahead of GPT-6 Astra, Gemini 4 Argon and Claude Fable 5.1 (53 each) and behind Claude Opus 5.5 (58). TechCrunch adds that Meta’s own model card shows scores comparable to Fable 5.1 and GPT-6 Astra, and that Muse Wonder leads on the MemoryBench memory benchmark in Meta’s testing.

Wonder Full’s standout is long-running coding (SWE-Marathon 57.5 vs 46.3 for Opus 5.5, per Meta’s model card), while rivals still lead some tests such as Terminal-Bench 4.0. The best test is your own tasks.

How much does Muse Wonder cost?

Through the Meta Model API (model id muse-wonder-1.0), Muse Wonder costs USD 4 per million input tokens and USD 20 per million output tokens, with cached input at USD 0.20, according to VentureBeat. That matches Claude Opus 5.5 per token and sits below GPT-6 Astra (USD 10/50). The Decoder notes it is about three times the price of the previous Muse Spark 1.3 model.

Wonder Full has no per-token API price. VentureBeat reports it is included in Muse Code’s Power Usage (USD 50/month) and Maximum Usage (USD 200/month) plans. These are Muse Code plans; they are separate from the Muse personal-agent plans listed on our Meta Muse page.

What does Muse Wonder mean for Muse users?

For people already using Meta Muse on WhatsApp, the iOS and Android apps or the web, the headline is simple: a stronger model is on the way, and subscribers get it first. Meta has not yet published a date or said which Muse plans count as “subscribers”. Better memory across conversations and more persistence on hard tasks map directly onto what personal agents are used for, such as planning and rebooking trips, a subscription audit or turning school emails into a daily plan.

Meta also describes safety training aimed at personal agents (Muse already runs each user’s agent in a dedicated Secure VM; see our Muse launch story): the models are trained not to share private information without permission, and a fuller safety report is promised “soon”.

What does it mean for businesses and developers?

Businesses get another frontier-class option for agents that work across documents and tools. Per VentureBeat, the API supports interfaces compatible with existing OpenAI and Anthropic SDKs, which makes side-by-side testing straightforward. Two details are worth planning around: the pay-as-you-go API runs in Meta’s “Standard mode” (submitted content not used for training), while Muse Code subscriptions default to a “Contributor mode” that users can switch off. And Wonder Full is not yet available for production API use.

If you are choosing between business agents rather than models, our comparisons of Grok Bot vs OpenAI Dots and OpenAI Dots vs Claude Cowork cover the agent layer, and business vs personal agents explains which kind fits which job.

Where does this leave the personal-agent race?

Muse Wonder arrives a month after Meta launched Muse and two weeks after OpenAI introduced Dots on GPT-6 Astra. Startups such as Instinct, Tab and Poke keep shipping too. For users this is good news: more capable models, more choice and pressure on price. See how Muse stacks up in OpenAI Dots vs Meta Muse, Instinct vs Meta Muse and our best personal AI agents 2026 guide.

Timeline: how did we get here?

Agent snapshot: what do we know so far?

Facts below come from the agent profiles in our directory, each linked to its source.

Meta Muse at a glance

Full Meta Muse profile

Related comparisons and guides

Agents in this story

Meta Muse

personal agent

Meta's personal AI agent: message it in the Muse app or WhatsApp, and it works on your goals inside a dedicated secure virtual machine.

Frequently asked questions

What is Meta Muse Wonder?

Muse Wonder is a family of two frontier AI models from Meta, Muse Wonder and Muse Wonder Full, launched on 12 October 2026 for personal and professional agent tasks such as coding, document work and long-running tasks.

Is Muse Wonder available now?

Yes. Muse Wonder is available through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. It will reach the Muse personal agent soon, first for subscribers.

How much does the Muse Wonder API cost?

USD 4 per million input tokens and USD 20 per million output tokens, with cached input at USD 0.20 per million (per VentureBeat). Wonder Full has no API price.

Is Muse Wonder open source?

No. Unlike Meta's earlier Llama models, neither Wonder model has downloadable weights, according to VentureBeat.

Will Muse Wonder power the Muse agent on WhatsApp?

Meta says Muse Wonder will soon power experiences in Muse, rolling out first to subscribers. It has not given a date.

Sources

Updated