What Is Muse Wonder?

Muse Wonder is Meta’s first frontier-class AI model, released on October 12, 2026 by Meta Superintelligence Labs. It comes in two sizes: Muse Wonder is available through the Meta Model API as muse-wonder-1.0 at $4 per million input tokens and $20 per million output tokens, and inside Muse Code, Meta’s terminal coding agent. Muse Wonder Full, the larger sibling, is available only to Muse Code subscribers on the two top plans; it has no API access and no published per-token price. Both models accept text, images, video, and audio, return text, and support a 1M-token context window.

Try Apidog today

Meta designed both models for agents: software that reads email, runs code, fills in forms, and keeps working after something breaks. That positioning shapes the entire release, from the two headline capabilities to the benchmarks Meta chose to publish. This guide covers what the models are, how they perform on Meta’s tests and independent evaluations, how to get access, what they cost, and where the catches are. If you want to call the model today, the Muse Wonder API guide walks through the first request in Apidog.

The Two Models at a Glance

	Muse Wonder	Muse Wonder Full
Model ID	muse-wonder-1.0	muse-wonder-full-1.0 (catalog only)
Where you get it	Meta Model API; Muse Code on all plans	Muse Code Power ($50/month) and Maximum ($200/month)
API price	$4 input / $20 output / $0.20 cached input per 1M tokens	None published
Context window	1M tokens	1M tokens
Inputs	Text, image, video, audio	Same
Output	Text	Text
Artificial Analysis Intelligence Index	52	56
Weights	Closed	Closed

Both models arrived about six weeks after Muse Spark 1.3, which scored 48 on the same index and costs $1.25 / $4.25 per million input/output tokens. Mark Zuckerberg described Wonder on X as “a much larger and more advanced class of models” than Spark. Meta’s research post says the models were trained at its gigawatt-scale Prometheus data center in Ohio, which came online this year.

What’s New: Active Memory and Active Focus

Meta highlights two capabilities that it says separate the Wonder series from earlier models.

Active Memory carries context about a person across conversations and retrieves the relevant parts for the current task. Meta frames this as model behavior rather than a separate memory API: there is no endpoint for writing memories directly. On Meta’s internal MemoryBench, which hides facts in 64K to 100K-token conversations and changes some facts midway through, Muse Wonder scores 85.6, compared with 81.5 for Claude Opus 5.5 and 81.4 for GPT-6 Astra.

Active Focus is Meta’s term for persistence toward a difficult goal: trying alternative approaches, working through blockers, adapting to new information, asking for clarification when needed, and checking in before important actions. On Meta’s internal FocusBench, Muse Wonder Full leads at 54.2%, followed by GPT-6 Astra at 47.7%, Claude Opus 5.5 at 45.8%, and the smaller Muse Wonder at 44.9%.

Both MemoryBench and FocusBench are internal Meta benchmarks and cannot currently be rerun outside Meta. Treat them as evidence of what Meta optimized for, then check the independent results below to see how that intent translates in practice.

Beyond those two headline features, Meta lists support for domain expertise in law, finance, and health; frontier coding across large repositories; computer use across browsers, desktops, and mobile apps; sub-agent delegation; parallel tool calls; and error recovery.

What the Benchmarks Say

Meta’s model card compares the two Wonder models with GPT-6 Astra and Claude Opus 5.5 at maximum reasoning effort. The selected tests favor agentic and professional work:

Benchmark	Wonder Full	Wonder	GPT-6 Astra	Claude Opus 5.5
Harvey LAB-AA (legal)	13.1	10.0	8.6	4.2
SWE-Marathon (long coding runs)	57.5	53.1	42.5	46.3
DeepSWE 1.1	79.0	78.8	74.1	74.2
JobBench	75.3	70.5	66.1	75.4
Terminal-Bench 4.0-AA	57.1	52.0	59.1	59.6
HealthBench Pro	62.3	60.5	65.3	65.6

The pattern is consistent: Wonder performs strongly on long-running agent work and legal tasks, while Opus 5.5 and GPT-6 Astra lead on terminal work and health tasks. Meta’s model card notes that competitor runs were best-effort and that some coding comparisons used different tools.

The independent view comes from Artificial Analysis. On its Intelligence Index v4.3, Muse Wonder Full scores 56, tied with Claude Sonnet 5.5, below Claude Opus 5.5 at 58, and above GPT-6 Astra, Gemini 4 Argon, and Claude Fable 5.1 at 53 each. Muse Wonder scores 52. Wonder Full’s largest gain over Muse Spark 1.3 was on Terminal-Bench 4.0, rising from 33% to 57%. Its hallucination rate fell from 33% to 23%, the second-lowest among frontier models. Knowledge recall remains a weakness: Wonder Full scores 30 on AA-Omniscience, compared with 46 for Opus 5.5. The Muse Wonder benchmarks post reviews every result, including one row where Meta’s table and Artificial Analysis disagree.

How to Get Access

Meta Model API

Create an API key at dev.meta.ai, point an OpenAI SDK-compatible client at https://api.meta.ai/v1, and set the model to muse-wonder-1.0. The API also serves Chat Completions and an Anthropic-compatible Messages endpoint, so tools such as Claude Code and Codex can use it with a few environment-variable changes. Rate limits are 3,000 requests and 4 million tokens per minute, shared across a team rather than per API key.

Muse Code

Muse Code is Meta’s terminal coding agent. It installs with a single curl command and includes Muse Wonder on every subscription tier from the $5 Everyday plan upward. The $50 Power and $200 Maximum plans unlock Muse Wonder Full. Subscriptions default to Contributor mode, which allows Meta to train on your prompts, code, and outputs in exchange for more usage; you can turn that setting off.

Muse App

The consumer Muse app is free, but Muse Wonder is rolling out first to its $20 and $100 subscribers. There is no free route to the model today. The free-access guide explains which options exist and what they cost.

Third-Party Gateways

Vercel AI Gateway and Cloudflare Workers AI both serve meta/muse-wonder-1.0 at the same token prices as Meta’s API.

What It Costs in Practice

At $4 input / $20 output per million tokens, Muse Wonder’s list price matches Claude Opus 5.5 and undercuts GPT-6 Astra and Claude Fable 5.1 ($10 / $50) by 60%. Cached input costs $0.20 per million tokens, a 95% discount. There is no long-context premium: a 900K-token prompt costs the same per token as a 9K-token prompt.

List price is not the same as task price. Artificial Analysis measured Muse Wonder at about $4.52 per Intelligence Index task, compared with $5.98 for Claude Opus 5.5 and $3.26 for GPT-6 Astra. The difference is token consumption: Muse Wonder used 7.69 million input tokens across the index, two to seven times more than the OpenAI and Google models evaluated, and most of its cost came from cache writes and reads rather than fresh input. A verbose agent on a cheap-looking rate can therefore still be expensive.

The Muse Wonder pricing guide works through practical cost examples. In Apidog, every response exposes usage.input_tokens_details.cached_tokens, so you can see exactly where your spend is going.

The Catches

• The strongest model is not available through the API. Meta’s frontier-performance claims are based on Muse Wonder Full, which is only available through paid Muse Code subscriptions. The API model scores four index points lower.
• Several benchmarks are internal. MemoryBench, FocusBench, and the Personal Superintelligence Index cannot be independently rerun. Meta also lists Wonder Full at 59.0 on an Artificial Analysis metric where Artificial Analysis publishes 56.
• Data modes matter. API traffic in Standard mode is not used for training. Muse Code subscriptions default to Contributor mode, which does permit training on prompts, code, and outputs. Zero data retention is available to approved organizations, but it disables server-side conversation state, background requests, file uploads, and built-in web search.
• Launch-week bugs appeared. Muse Wonder and Muse Wonder Full drained Muse Code quotas abnormally quickly on launch day. Meta reset affected quotas, and Alexandr Wang said there were “still issues we’re trying to iron out.”
• The safety report is still pending. Meta says it could not rule out crossing the high-risk threshold for chemical and biological misuse without safeguards. With safeguards, it assessed residual risk as moderate or lower and promised a full report soon.
• Agents still make mistakes. Meta recommends human confirmation before consequential actions—the same precaution to apply to any agent that can interact with your APIs.

Frequently Asked Questions

Is Muse Wonder open weights like Llama?

No. Both Muse Wonder models are closed and served only through Meta’s API, Muse Code, and partner gateways. There are no downloadable weights.

Which API format does Muse Wonder use?

It supports three API surfaces: the Responses API, which Meta recommends for multi-step work; Chat Completions, which works with OpenAI SDKs; and an Anthropic-compatible Messages endpoint. Reasoning effort ranges from minimal to max. Setting it to none returns a 400 error because the model always reasons.

Is it better than Claude Opus 5.5?

It depends on the workload. On Meta’s agentic benchmarks, Muse Wonder Full leads on long coding runs and legal work. On Artificial Analysis’ independent index, Claude Opus 5.5 remains ahead of both Wonder models and also performs better on terminal and health tasks. The head-to-head comparison covers the details; the short version is that they are priced identically and win different jobs.

Can I try it before paying?

Only by spending cents on the API. A few test calls at $4 / $20 per million tokens usually cost less than a coffee. Download Apidog, store your API key as a secret variable, and run your test requests from one project.

Where This Leaves You

Muse Wonder is a credible frontier model priced like Claude Opus 5.5, with a clear advantage on long-running agent tasks and clear gaps in knowledge recall and terminal work. The version behind Meta’s headline benchmark results requires a $50 Muse Code subscription. If you build agents, test Muse Wonder against your own tasks this week—and measure tokens consumed, not just list price.