Muse Wonder Full Pricing: What You Actually Pay to Use It
Muse Wonder Full costs $50 or $200 a month through Meta's subscription tiers, with no published pay-as-you-go API rate yet.
How much does Muse Wonder Full cost?
Muse Wonder Full is available through two Meta subscription tiers at launch pricing: a power usage plan at $50 a month and a maximum usage plan at $200 a month. There is no published pay-as-you-go API rate for Full. Standard Muse Wonder, by contrast, does have a listed API price ($4 per million fresh input tokens, 20 cents per million cached input tokens), which means anyone trying to estimate Full’s per-token cost is working from an assumption, not an official number.
TL;DR
- Full only ships via subscription right now, locked behind the $50 power usage tier or the $200 maximum usage tier, with no self-serve API price sheet to point to.
- Standard Wonder has a public API rate ($4/million fresh input tokens, $0.20/million cached), but that rate does not officially extend to Full.
- Selecting “maximum reasoning” on standard Wonder is not the same as using Full. They are separate model variants, and mixing them up will give you a false read on cost and capability.
- Any dollar estimate for Full’s token cost is borrowed math, applying standard Wonder’s published rates to Full as a stand-in, not a real quoted price.
- Full scores higher than standard Wonder on several public benchmarks (Val index, Terminal Bench, Vibe Code Bench, Artificial Analysis intelligence and coding indices), which is the usual justification Meta and third parties offer for the pricing gap.
- The weights are not public, so there’s no self-hosted or open-weight path to sidestep the subscription cost, unlike some other model families.
- Real-world task performance doesn’t track cleanly with the higher price or higher benchmark scores, based on independent testing across coding and app-building tasks.
Other agents ship a demo. Remy ships an app.
Real backend. Real database. Real auth. Real plumbing. Remy has it all.
What’s the difference between the $50 and $200 tiers?
Meta’s launch pricing frames the two tiers as “power usage” ($50/month) and “maximum usage” ($200/month). The names suggest the split is about usage volume or rate limits rather than feature gating, consistent with how most frontier labs structure consumer-facing subscription tiers: a mid tier for regular use and a higher tier for heavier or more frequent workloads. Beyond that framing, Meta hasn’t published a detailed breakdown of what changes between the two tiers (token caps, concurrent task limits, context window differences, or priority access), so builders evaluating which tier to pick are largely going on the tier names and price points alone.
This matters for anyone budgeting for a coding agent or app-building workflow. A $50/month plan that throttles you after moderate use is a very different value proposition than a $200/month plan with headroom for daily heavy use. Without published usage caps, the practical difference between the tiers is something you’d need to discover by hitting limits in actual use.
Why doesn’t Muse Wonder Full have a published API price?
Full is a separate model variant from standard Wonder, and Meta has not released a pay-as-you-go tariff for it. Standard Wonder does have one: $4 per million fresh input tokens and $0.20 per million cached input tokens. Some evaluations and cost estimates apply those same standard-Wonder rates to Full as a placeholder, explicitly flagging the number as an assumed API equivalent, not an official price. That distinction matters because it means any “cost per task” or “cost per benchmark run” figure you see for Full elsewhere is likely built on the same borrowed assumption, not a number Meta has actually billed anyone.
The practical effect is that Full’s cost structure is currently locked into subscription seats rather than metered consumption. If you want to use Full at all today, you go through the $50 or $200 monthly plan. There isn’t yet a developer-facing way to pay only for the tokens you consume, which puts Full in a different category from most API-first model releases where pay-as-you-go is the default on-ramp and subscriptions come later for power users.
Is Muse Wonder Full worth the subscription price?
The honest answer depends on what you’re using it for, and the benchmark and real-world task evidence available so far is mixed.
On public benchmarks, Full generally outperforms standard Wonder. In one index tracked around the model’s launch, Full placed fifth at roughly 64.57% versus standard Wonder’s 10th-place 61.06%. On a terminal-focused coding benchmark, Full landed around fifth place with a score near 58%. On a “vibe code” benchmark, Full scored about 86% (12th place) against standard Wonder’s 83% (18th place). Artificial Analysis’s intelligence index gave Full a score of 56 versus 52 for standard Wonder, and its coding agent index (which measures the model paired with Meta’s Muse code tool) gave Full a 68 versus 63 for standard Wonder. Not every benchmark favored Full: on a code migration test, standard Wonder actually scored slightly higher (62.55% vs. 61.03%).
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
Independent hands-on testing across six practical app-building tasks (an emulator, a Docker alternative, a 3D room designer, a keyboard configurator, a Pokémon-style game, and a desktop app) found a more uneven picture. Full beat standard Wonder clearly on some tasks, like the retro game emulator (3.1/10 vs. 2.4/10) and the keyboard configuration tool (5.5/10 vs. 3.5/10), where it implemented more of the connected workflow even though protocol-level details and error handling still had real bugs. On the Docker alternative, both models tied at 3.0/10, capped by the same category of storage and integrity failures. On the 3D room designer, Full actually scored lower than standard Wonder (7.00 vs. 7.6), despite producing a more visually complete interface, because of editing bugs, broken undo behavior, and resource leaks that standard Wonder didn’t have.
That pattern (higher benchmark scores, inconsistent task-level results) is worth keeping in mind before treating Full’s price premium as a straightforward trade of “pay more, get proportionally better output.” For buyers weighing the $50 versus $200 tier, the decision probably comes down to usage volume and whether the kinds of tasks you run tend to land in the categories where Full’s improvements actually showed up.
How does Full’s lack of open weights affect cost?
Muse Wonder and Muse Wonder Full are both proprietary. The weights are not publicly available, so there is no option to download the model and run it on your own hardware to avoid subscription or API costs. That’s a meaningful difference from open-weight model families, where a determined builder can sidestep ongoing fees entirely by self-hosting. With Full, the subscription (or, eventually, a metered API) is the only path to using the model at all.
Frequently Asked Questions
What are the two Muse Wonder Full subscription tiers?
Meta’s launch pricing lists a power usage tier at $50 a month and a maximum usage tier at $200 a month. Both grant access to Full; the exact usage limits separating the two tiers haven’t been published in detail.
Does Muse Wonder Full have pay-as-you-go API pricing?
No. Standard Muse Wonder has a published API rate ($4 per million fresh input tokens, $0.20 per million cached). Full does not have an official pay-as-you-go tariff; any per-token cost estimate for Full is an assumption based on standard Wonder’s rates, not a confirmed price.
Is Muse Wonder Full the same as standard Wonder with maximum reasoning turned on?
No. They’re separate model variants. Selecting the highest reasoning setting on standard Wonder does not switch the underlying model to Full. You have to explicitly select Full (for example, through Meta’s native CLI) to use it.
Can I run Muse Wonder Full locally to avoid subscription costs?
No. The model weights for both Muse Wonder and Muse Wonder Full are not publicly released, so there’s no self-hosted alternative to the subscription or eventual API access.
Is the $200 tier worth it over the $50 tier?
That depends on usage volume, since Meta hasn’t detailed exactly what separates the two plans beyond their price and tier names. Heavy, frequent users of coding or agentic workflows are the likeliest candidates for the $200 tier, but anyone evaluating the jump should weigh it against actual task-level performance rather than benchmark scores alone, since independent testing shows Full’s real-world gains over standard Wonder vary significantly by task.
