Muse Wonder is Meta's most capable AI model, released on October 12, 2026 alongside a stronger sibling called Muse Wonder Full. The short version: Muse Wonder is available now to developers through the Meta Model API at $4 per million input tokens and $20 per million output tokens, and inside Meta's Muse Code coding tool. Muse Wonder Full is not sold through the API at all; you only get it on Muse Code's $50 or $200 monthly plans. Meta says Wonder will soon power its Muse personal agent too, starting with paying subscribers.
This guide explains what Meta has actually confirmed: the two variants, where each one is available, what it costs, how it differs from the cheaper Muse Spark line, and how Meta positions it against Claude, GPT-6 and Gemini. Facts come from Meta's Muse Wonder research post, the Meta developer launch post and Meta's model documentation, cross checked against independent reporting. Benchmark claims are Meta's own unless we say otherwise. The page reflects the situation on October 16, 2026.
What is Muse Wonder?
Muse Wonder is a large multimodal model from Meta's Superintelligence Labs, built for what Meta calls "personal and professional agentic tasks": long coding jobs, multistep work across apps, and assistant tasks that need to remember context over time. Meta describes the Wonder series as "the next step on our scaling ladder and a major leap toward personal superintelligence," the phrase Mark Zuckerberg's company uses for its long term goal. TechCrunch called it Meta's first attempt at a true frontier model, since the earlier Muse Spark line was designed to be lighter and faster.
A few confirmed details:
- Context window: one million tokens, according to both the developer launch post and Meta's models documentation.
- Inputs: text, images, video, audio and PDF in, text out, on the same endpoints as Muse Spark.
- Training: Meta says both Wonder models were trained at its gigawatt scale Prometheus data center, which TechCrunch locates in central Ohio.
- Model ID: developers call it as
muse-wonder-1.0.
Meta highlights two headline abilities. Active Memory is the model carrying important context about you across conversations and pulling up the right detail when it becomes relevant. Active Focus is its persistence on hard tasks: trying another approach when one fails, working through blockers, asking for clarification when needed and checking in before important actions. Both are aimed squarely at agents that run for a long time on a person's behalf.
Muse Wonder vs Muse Wonder Full
The two names cause most of the confusion, so here is the split in one place.
| Item | Muse Wonder | Muse Wonder Full |
|---|---|---|
| What it is | The standard Wonder model | Meta's higher capability variant |
| Meta Model API (pay per token) | Yes, $4 in / $20 out per 1M tokens | Not available |
| Muse Code | Every plan and pay as you go | Power Usage ($50/mo) and Maximum Usage ($200/mo) only |
| Context window | 1M tokens | 1M tokens |
| Coming to the Muse agent | Yes, subscribers first | Meta has not said |
Meta's developer post says Muse Wonder "brings comparable performance to Muse Wonder Full," while Full is "our most advanced model yet." One detail matters for developers: setting the reasoning effort to its maximum on the API does not quietly switch you to Full. Meta's documentation says so explicitly, and TechCrunch also reported that Full "will not be served through the API."
Muse Wonder pricing and API access
On the Meta Model API pricing page, Muse Wonder costs:
| Usage | Price per 1M tokens |
|---|---|
| Input | $4.00 |
| Cached input | $0.20 |
| Output | $20.00 |
In plain terms, that is four dollars per million tokens you send, twenty dollars per million tokens the model writes back, and twenty cents per million for prompt text served from cache.
Meta's cheaper "Contributor mode," where you pay less in exchange for letting Meta train on your prompts and outputs, is not offered for Muse Wonder on the pay as you go API. Wonder runs only in Standard mode there, and Meta says Standard mode content is not used to train its models.
The API accepts OpenAI style requests (Responses and Chat Completions) as well as an Anthropic style Messages endpoint, so existing apps can often switch by changing the base URL, key and model ID. If API keys and per token billing are new to you, our explainer What Is an LLM API Key? covers the basics.
Muse Code plans: where Wonder Full lives
Muse Code is Meta's terminal coding agent, and it is the only place to use Wonder Full. Its subscription plans, from the Muse Code product page and the subscriptions documentation:
| Plan | Price | Usage | Wonder Full? |
|---|---|---|---|
| Everyday Usage | $5/month | 10 to 50 prompts every 5 hours | No |
| High Usage | $15/month | 5x Everyday | No |
| Power Usage | $50/month | 20x Everyday | Yes |
| Maximum Usage | $200/month | 80x Everyday | Yes |
Every plan includes the standard Muse Wonder. The Maximum Usage plan is new with this launch. Two catches: subscriptions default to Contributor mode, which gives you more usage in exchange for letting Meta train on your content (you can switch it off in settings), and you must be 18 or the age of majority in your country to subscribe.
Muse Wonder in the Muse app
For everyday users, Wonder's main destination is Muse, the personal AI agent Meta launched in the US on September 8, 2026. Muse connects to apps such as email, calendars and payments to take actions for you, and is available on the web, on iOS and Android, and in WhatsApp chats, as TechCrunch reported at launch.
Muse itself is free with a usage limit. Meta's Muse subscriptions help page lists a Power plan at $20 a month and a Maximum plan at $100 a month for higher weekly usage. Meta says Muse Wonder "will soon power experiences in Muse, and will roll out first to subscribers," and Reuters reported the same. Meta has not given a date.
Note that Muse is not the same thing as the free Meta AI assistant inside Facebook, Instagram and WhatsApp. Meta's Wonder announcements name Muse, Muse Code and the API, not Meta AI in its social apps. For how Meta AI compares as a free chatbot, see Meta AI vs ChatGPT.
Muse Wonder vs Muse Spark
Muse Spark is Meta's earlier, lighter model line, and it is still on sale. Meta's documentation says the two take the same requests on the same endpoints, so switching is just a model ID change.
| Item | Muse Wonder (1.0) | Muse Spark 1.3 |
|---|---|---|
| Input / output per 1M tokens (Standard) | $4.00 / $20.00 | $1.25 / $4.25 |
| Contributor mode on the API | Not available | Yes, $0.10 / $0.20 |
| Best for, per Meta | Long, complex tasks: multifile coding, multihour agent runs | High volume, fast or cost sensitive work |
| Open weight version | No | Muse Spark Open (released Oct 15) |
Meta's guidance is to use Wonder when capability matters more than cost per token, and Spark for subagent tasks, classification and tight tool loops. Meta also says Wonder Full handles unfamiliar tasks better than Muse Spark 1.3's trend would predict, because the Wonder models build on larger, stronger base models.
People searching "Muse Wonder open" may be thinking of Muse Spark Open, whose weights Meta released on October 15. Muse Wonder itself has no downloadable weights; it is only available through Meta's own services.
How Muse Wonder compares with Claude, GPT-6 and Gemini
Meta claims state of the art results on agent, memory and coding benchmarks. Its developer post says Wonder Full tops the DeepSWE and SWE-Marathon coding leaderboards "in independent third party runs." TechCrunch, summarizing Meta's model card, reported that Wonder and Wonder Full score comparably to Anthropic's Claude Fable 5.1 and OpenAI's GPT-6 Astra, and that Meta's testing puts Wonder ahead of Fable, Opus and Astra on the MemoryBench memory test. Treat these as vendor claims until more independent testing is published.
On price, Muse Wonder matches Anthropic's Claude Opus 5.5, which Anthropic's models overview also lists at $4 input and $20 output per million tokens. Claude Fable 5.1 costs more, at $10 and $50. For the full Anthropic lineup see Claude Fable vs Opus vs Mythos, and for Google's newest flagship see Gemini 4 Argon release date.
The bigger difference is the product around the model. Claude, ChatGPT and Gemini all put their top models inside mature consumer chat apps. Meta is routing its best model to developers first and to Muse subscribers later, and its very best variant is locked to one coding tool.
Safety and the personal superintelligence pitch
Because a personal agent may read messages, calendars and private documents, Meta says it trained the Wonder family to resist adversarial instructions and handle private information carefully. It says Wonder has its strongest safeguards yet against chemical, biological, radiological, nuclear and cyber misuse, under its Meta Superintelligence Scaling Framework, while keeping defensive cyber capabilities available. Meta calls Wonder "our most aligned model to date" and says a full report with additional safety evaluations "will be available soon." For the lab behind the model and what Meta means by personal superintelligence, see our Meta Superintelligence Labs explainer.
If you want to understand how lab safety frameworks work in general, our comparison of OpenAI's Preparedness Framework and Anthropic's Responsible Scaling Policy is a good primer.
Pros and cons
Pros
- Priced the same as Claude Opus 5.5 on the API, with cheap cached input.
- One million token context and multimodal input such as screenshots, documents and video.
- Drop in compatible with OpenAI style and Anthropic style API requests.
- Built around long running memory and persistence, which suits agents.
- Prompts in Standard mode are not used for training.
Cons
- The strongest variant, Wonder Full, is only in Muse Code and never on the API.
- No free way to use it yet; Muse Code has no free plan and the Muse rollout has no date.
- Benchmark claims are mostly Meta's own so far, and the full safety report is not published.
- Muse Code subscriptions default to letting Meta train on your content.
Who should try Muse Wonder
- Developers building agents or long coding workflows: try
muse-wonder-1.0on the API and compare it with Opus 5.5 on your own tasks at the same price. - Developers who want the very top Meta model: Wonder Full means a $50 or $200 Muse Code plan.
- Cost sensitive, high volume apps: start with Muse Spark 1.3, which costs much less per token.
- Everyday users: wait for Wonder to reach Muse, or stick with the consumer apps you already use.
FAQ
Is Muse Wonder free?
Not right now. On the API it is pay per token, at $4 per million input tokens and $20 per million output tokens. In Muse Code, the cheapest plan with Muse Wonder is Everyday Usage at $5 a month. Meta says Wonder is coming to its Muse agent soon, but paying Muse subscribers get it first, and no date has been given.
Can I use Muse Wonder Full through the API?
No. Meta's documentation and TechCrunch both say Wonder Full is only available in Muse Code on the Power Usage ($50) and Maximum Usage ($200) plans. Maximum reasoning effort on the API does not select it.
Is Muse Wonder open source?
No. Muse Wonder is only available through Meta's API and products. Meta's open weight release from the same week is Muse Spark Open, a different model.
Is Muse Wonder cheaper than Claude Opus 5.5?
They cost the same on their APIs: $4 per million input tokens and $20 per million output tokens. Muse Wonder's cached input is $0.20 per million tokens. Which is better value depends on how many tokens each needs for your task, so test both.
Does Meta train on my Muse Wonder prompts?
On the pay as you go API, Muse Wonder runs only in Standard mode, where Meta says your content is not used for training. In Muse Code subscriptions, Contributor mode is on by default and lets Meta train on your content in exchange for more usage; you can turn it off in settings.
When did Muse Wonder come out?
Meta released Muse Wonder and Muse Wonder Full on October 12, 2026. Muse Wonder was available the same day on the Meta Model API and in Muse Code, and Wonder Full for Muse Code subscribers.
