Meta Muse Wonder Launch: Price, Access and Claims

Quick answer: On 12 October 2026 Meta launched Muse Wonder and Muse Wonder Full, which it calls its first frontier-class AI models. Muse Wonder is available through Meta’s API at $4 per million input tokens and $20 per million output tokens, while Wonder Full is limited to paid Muse Code subscribers. The benchmark claims are mostly Meta’s own.

Meta announced the two models on Monday, 12 October, after months of work at its Superintelligence Labs. They were previously code-named Watermelon, according to Axios. Meta says they are state of the art for personal and professional agents, memory and coding.

Most everyday users cannot try them directly yet. The models are aimed at developers and paying subscribers first, and Meta says Muse Wonder will power future experiences in its Muse assistant, initially for paid users. The details below apply globally unless a source says otherwise; none of the sources spell out country-by-country availability.

What did Meta actually launch?

Two models: Muse Wonder and Muse Wonder Full. TechCrunch describes them as larger and more capable than the lightweight Muse Spark, with better results on complex tasks such as coding. Meta’s blog calls them “the next step on our scaling ladder and a major leap toward personal superintelligence.”

Meta says they were trained at its Prometheus data center. Two features get the most attention:

Active Memory: keeps information about a user across conversations and retrieves it when relevant.
Active Focus: helps an agent keep working toward a goal, trying different approaches when it hits obstacles.

How much does it cost and who can use it?

VentureBeat reports the following for the API model, ID muse-wonder-1.0:

Input: $4 per million tokens. Output: $20 per million tokens. Cached input: $0.20 per million.
Web search: $2.50 per 1,000 queries.
Context window of about one million tokens; text, image, audio and video input with text output.

Muse Wonder Full has no announced pay-as-you-go price. It is available only through Muse Code subscriptions, reported at $50 per month (Power Usage) and $200 per month (Maximum Usage), both with usage limits. Neither model has public weights, so neither can be downloaded or self-hosted. Prices for UK, EU, Canadian or Australian customers were not given in the sources.

How does it compare with rivals?

VentureBeat cites the Artificial Analysis Intelligence Index: Claude Opus 5.5 scores 58, Wonder Full 56, GPT-6 Astra, Gemini 4 Argon and Claude Fable 5.1 score 53, and Wonder scores 52. Quartz adds that Wells Fargo said Meta has reached the “AI frontier” with Wonder Full, and that Meta shares rose about 3% on Tuesday after a two-day gain of roughly 6%.

Treat the rest with care. TechCrunch notes Meta’s benchmark results come from its own model card, with no independent verification. VentureBeat adds that Meta’s comparison table mixes internal runs, vendor scores and third-party results. For a plain-language look at how today’s main assistants differ, see our guide to ChatGPT vs Gemini vs Claude.

What should users know about privacy?

VentureBeat reports that Muse Code subscriptions default to a Contributor mode that may let Meta use prompts, code, files and outputs to improve models. Turning it off applies only to future submissions, and Meta’s terms say not to submit confidential or personal information in that mode. The API’s Standard mode is stated not to train on submitted content.

Memory features raise their own question. Axios lists as open issues whether the performance claims hold up in independent tests, how fast the technology reaches Meta’s users, and the privacy and security risks of agents that remember personal information. If you use any assistant with memory, review what it stores; our guide on how to stop AI chatbots training on your data walks through the settings.

What is still unclear?

Whether base Muse Wonder reaches consumers beyond the Muse agent, and what Muse Code subscriptions cost outside the reported US dollar tiers (TechCrunch says it does not say).
Whether Meta will open-source any Wonder models. Axios says Meta has not said, although Meta’s Alexandr Wang stated, “We’re still committed to open source.”
How the models perform in independent testing on unfamiliar multi-step tasks, where Meta itself acknowledges mistakes can happen and recommends human confirmation for consequential agent actions.

Why does this launch matter?

TechCrunch calls it the first frontier release since Alexandr Wang restructured Meta’s AI efforts in August 2025. Until now Meta had released the lightweight Muse Spark and the Muse agent without a comparable state-of-the-art model. The move also tests whether people will pay for Meta’s AI: the Muse Code tiers sit alongside the plans from OpenAI, Google and Anthropic, and Axios notes Meta says the models do well on professional work in law, finance and health. Those are Meta’s claims, so do not treat any AI output on such topics as advice.

Meta also says it strengthened safeguards against cyber and biological misuse and added protections for personal information. A model card was due on launch day, with a fuller report to follow, according to Axios. Readers who want detail should wait for that report and independent reviews.

Follow the latest launches and rules in our more AI coverage section.

Frequently asked questions

What is Muse Wonder?

It is Meta’s new frontier-class AI model, launched 12 October 2026 alongside a more capable version called Muse Wonder Full.

How much does the Muse Wonder API cost?

VentureBeat reports $4 per million input tokens and $20 per million output tokens, with web search at $2.50 per 1,000 queries.

Can I download Muse Wonder?

No. Neither model has public weights, according to VentureBeat, so they cannot be self-hosted.

Is Muse Wonder better than Claude or ChatGPT?

Meta’s own benchmarks say it matches or tops rivals in places, but on the Artificial Analysis index cited by VentureBeat, Claude Opus 5.5 still scores higher than Wonder Full. Independent testing is still limited.