Muse Wonder: Pricing, Full & API (2026)

Muse Wonder pricing starts at $4 per 1M input tokens on Meta Model API, while Muse Wonder Full needs a $50 or $200 Muse Code plan. See each plan and limit.

Released on October 12, 2026, Muse Wonder and Muse Wonder Full are Meta’s most capable models so far, built for agents that work through long tasks rather than one-off answers. This page explains how the two models differ, what each one really costs, where you can use them today and what Meta’s own benchmark table does and doesn’t prove.

What is Muse Wonder?

Muse Wonder is a frontier reasoning model from Meta Superintelligence Labs and the successor to Muse Spark. Meta pitches it at personal and professional agent work: navigating an insurance claim, reading a lease, untangling finances or building a working app. It accepts text, images, video, audio and PDF files, returns text and has a context window of about 1 million tokens, according to its OpenRouter listing.

Two features stand out in Meta’s description. Active Memory carries the details that matter across conversations, and Active Focus keeps the model working toward a goal when a task gets hard, trying other approaches and checking in before anything important. Meta also says it plans long jobs, hands parts to sub-agents, calls tools in parallel and can operate browsers, desktop and mobile apps.

Don’t confuse the model with the app. Meta Muse is Meta’s personal AI agent for consumers, and it currently runs on Muse Spark. Meta says Muse Wonder will soon power more of Muse, starting with paying subscribers.

Muse Wonder vs Muse Wonder Full

• Muse Wonder (muse-wonder-1.0): the model developers can call today on Meta Model API, on OpenRouter and inside Muse Code. It has a public per-token price.
• Muse Wonder Full (muse-wonder-full-1.0): Meta calls it its most advanced model yet. It has no per-token price and isn’t offered on the API. You get it only through a Muse Code subscription on the Power or Maximum plan.
• Same context window: both are listed with a 1M-token context on Meta’s developer site.

In practice, Wonder is the model you build products on, while Wonder Full is a perk for developers who pay for Meta’s coding agent.

Muse Wonder Pricing (2026)

We checked Meta’s Muse Wonder model page and Muse Code plans on October 14, 2026. Meta’s developer site is region-restricted and showed us an unavailable notice from our location, so we confirmed the API rates against the archived page from October 13 and the live OpenRouter listing, which match.

Muse Wonder on the API	Input	Cached input	Output	Context
Meta Model API (muse-wonder-1.0)	$4.00 per 1M tokens	$0.20 per 1M tokens	$20.00 per 1M tokens	1M tokens
OpenRouter (meta/muse-wonder)	$4.00 per 1M tokens	$0.20 per 1M tokens	$20.00 per 1M tokens	1M tokens, plus web search at $2.50 per 1,000 queries

Plan	Price	Muse Wonder	Muse Wonder Full	Usage
Muse Code Everyday	$5/month	Yes	No	10 to 50 prompts every 5 hours
Muse Code High	$15/month	Yes	No	5x Everyday
Muse Code Power	$50/month	Yes	Yes	20x Everyday, early features
Muse Code Maximum	$200/month	Yes	Yes	80x Everyday
Muse app Power / Maximum	$20 / $100 a month	Coming, subscribers first	Not announced	500M / 3B Muse tokens a week

What does that mean in money? A request with 100,000 input tokens and 10,000 output tokens costs about $0.60 on the API ($0.40 in, $0.20 out). If the 100,000 input tokens are cached, the input part drops to $0.02. That caching discount matters for agents, which resend the same long context many times.

For comparison, OpenRouter lists Claude Opus 5.5 at the same $4 and $20, while GPT-6 Astra and Claude Fable 5.1 cost $10 and $50 per 1M tokens. So Muse Wonder isn’t a budget model; it’s priced like Anthropic’s Opus tier. See our GPT-6 Astra vs Sol breakdown for OpenAI’s side.

Is Muse Wonder free? Not in any confirmed way. There’s no free API tier, and Meta says Wonder will reach the Muse app first for subscribers. Reports that free Muse users already get Wonder aren’t confirmed by Meta, so plan on paying.

Muse Wonder benchmarks: Meta’s own numbers

Meta published a comparison table on the model page. These are Meta’s figures, not an independent test, though Meta says its DeepSWE and SWE-Marathon results come from third-party runs. All models were scored at their highest reasoning setting.

Benchmark	Wonder Full	Wonder	Claude Opus 5.5	GPT-6 Astra	Claude Fable 5.1
DeepSWE 1.1 (long coding)	79.0	78.8	74.2	74.1	67.4
SWE-Marathon	57.5	53.1	46.3	42.5	45.6
Terminal-Bench 4.0-AA	57.1	52.0	59.6	59.1	52.0
ProgramBench	59.0	50.0	70.5	52.5	55.5
OSWorld V2.1 (partial)	78.4	72.6	81.8	75.8	80.7
Artificial Analysis Index	56	52	58	53	53

Our read: Wonder Full leads the long-horizon coding tests Meta highlights, but Claude Opus 5.5 still wins Meta’s own table on terminal coding, programming from scratch, computer use and the Artificial Analysis Index. The cheaper Wonder you can buy per token trails Opus 5.5 on most rows at the same price. Check our best AI model guide before switching.

How to use Muse Wonder

1. Meta Model API: point an OpenAI SDK-compatible client at Meta Model API and use the model name muse-wonder-1.0. Meta opened the API as a preview for US developers.
2. OpenRouter: call meta/muse-wonder with your OpenRouter key. This is the easiest route if Meta’s developer site isn’t available where you are.
3. Muse Code: install Meta’s coding agent for macOS or Windows and pick a plan; Power or Maximum unlocks Wonder Full.
4. Muse app: Meta says Wonder is coming to Muse soon, starting with subscribers.

Limitations

• Wonder Full is subscription-only: no API access and no per-token price.
• Region limits: Meta’s developer site and Muse aren’t available everywhere.
• Benchmarks are self-reported: independent results are still thin.
• Not cheap: same list price as Claude Opus 5.5, and output tokens cost five times input.
• No free access confirmed: free Muse users haven’t been promised Wonder.

Muse Wonder alternatives

• Claude Sonnet 4.5: Anthropic’s mid-tier model for coding and agents at a lower price.
• GPT-6 Astra and Sol: OpenAI’s flagship pair, with Sol much cheaper than Astra.
• Meta Muse: Meta’s consumer agent, free to start, if you want tasks done rather than an API.

Muse Wonder puts Meta in the frontier race for agent and coding work, and the 1M context with cheap cached input suits long-running agents. At $4/$20 it is priced like Claude Opus 5.5, and on Meta’s own table Opus 5.5 still wins more rows.

Try it through OpenRouter on a real task before switching. Pay for a $50 Muse Code plan only if Wonder Full’s long-horizon coding results matter to your work.