BLOG
Meta announced Muse Wonder and Muse Wonder Full on October 12, 2026. The choice depends on the job: Wonder is offered through the Model API, while Full is available through Muse Code’s subscription plans. The published prices for API tokens and monthly software plans describe different products. Check account eligibility, usage limits, data settings, and the current bill before shifting a development workflow. This article uses Meta’s public documentation; we did not run our own model benchmark.
What Meta announced on October 12
Meta introduced Wonder as a model aimed at agentic and coding work and Wonder Full as a more capable option inside Muse Code. The launch announcement and model page describe these offerings. API access and a coding subscription do not automatically share one account balance, quota, or billing method.
Documentation can change during a preview. Confirm the exact model ID, current access status, and account limits in the developer console. A vendor’s demonstration or benchmark is a reason to investigate, not proof that the model will outperform another one on your repository. Language, codebase, task design, tool access, and review criteria all affect the result.
How the Model API is priced
Meta’s pricing documentation lists muse-wonder-1.0 in Standard mode at $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. The model page lists a context window of up to one million tokens. These are published USD rates, not the cost of a completed task. Check the latest prices, tax treatment, limits, and account region before approving a budget.
A large context window does not mean every call should send the whole repository. Unnecessary files increase exposure and can raise cost. Start with relevant code, retrieval, and concise instructions. Estimate total spend from observed input, cached input, output, number of calls, retries, and tool use. Set spending alerts and a pilot ceiling. Compare providers by the amount of reviewed work completed, not only by nominal token price.
Muse Code is a separate subscription decision
Muse Code is Meta’s development experience. Wonder Full is described as available through its Power Usage and Maximum Usage subscription plans. The launch material announced monthly plan prices of $50 and $200 respectively; confirm the live plan page for current amounts, eligibility, taxes, seats, and usage terms. Do not treat a Muse Code subscription as general-purpose credit for the Model API.
Read which models are included in each plan and how quotas or rate limits work. For a team, check administrative controls, seat allocation, data retention, and contribution settings. If you need calls from your own app or an n8n workflow, assess the API and its variable billing. If you need an interactive coding environment, assess Muse Code against your daily editor and review process.
| Decision | Model API | Muse Code |
|---|---|---|
| Typical use | Integrate model calls into software | Interactive coding work |
| Billing to verify | Usage, tools, and token rates | Plan, seats, and quotas |
| Data controls | API mode and project settings | Subscription contribution setting |
| Operational test | Backend security and cost ceiling | Repository workflow and review |
Preview status and regional access
The model documentation and current console should determine whether your project has access. Public preview may have different eligibility, quotas, stability, or support from a broadly available product. Do not interpret a published landing page as proof that every account can activate the model. Check project creation, country rules, payment method, waitlist status, and the specific model selector.
For a US business, verify the state of your own account rather than assuming access from the national market. If the model is missing, the cause could be eligibility, rollout, organization policy, or account restrictions. Do not work around location requirements with false information. Keep a supported alternative and a provider boundary in the application while access is uncertain.
Read the data and contribution settings first
Meta’s pricing and mode guide distinguishes Standard from Contributor mode. It says Standard mode content is not used to train Meta models. The same page says Contributor mode can use submitted prompts, code, files, context, and responses for future training in exchange for discounted access where offered. It also states that Contributor mode is not available for Muse Wonder through the Model API. Muse Code subscriptions default to Contributor mode but offer a setting to turn it off. Verify the setting in the account before sending private code.
Test initially with a public repository or synthetic project. Remove tokens, .env files, SSH keys, customer records, and production dumps. A team should approve the applicable retention and training terms, document the selected mode, and recheck settings after a new login or policy update. Regulated or confidential data requires its own legal and security review; a model’s technical ability to accept a file is not permission to upload it.
Run a fair coding pilot
Choose five tasks from a non-sensitive project: explain a module, locate a bug, propose a small change, create a test, and refactor a function. Use the same code version, task instructions, and acceptance criteria for each tool. Record whether the project builds, which tests pass, manual corrections, elapsed time, cost, and security concerns. A handpicked vendor demo cannot replace this comparison.
Use a disposable branch with no production credentials. Inspect every diff, dependency, test, and license-sensitive addition. Repeat important tasks to see whether results vary, without calling a small pilot a universal benchmark. Record the exact model, plan, and data mode; Wonder and Wonder Full are different offerings. If a subscription hits a quota, record the interruption rather than assuming its economics match API usage.
Match the option to the team
A software team embedding AI in a product needs API access, server-side key storage, request limits, logging, and variable-cost controls. An individual developer seeking help inside a dedicated coding tool may evaluate Muse Code, including monthly cost and review workflow. A small business without a developer should first define the task, owner, and expected output: buying a coding subscription does not create a CRM integration on its own.
For tasks that require predictable business outcomes, combine model output with deterministic validation and human escalation. For private code, evaluate confidentiality as a procurement requirement. If an account cannot activate the selected product, do not plan production around it. The suitable option depends on access, total cost, data policy, and observed quality in your own tasks; the announcement does not establish a universal winner.
Checklist before committing
Confirm the date and current status; model ID; project and regional eligibility; API versus subscription; currency and tax; context and rate limits; data retention and training mode; opt-out control; team management; spending alerts; test environment; and fallback provider. Keep links and the date of verification because preview documentation changes.
Run a limited pilot with an approved repository and a firm spending ceiling. Check the invoice, usage logs, and data controls afterward. In Muse Code, confirm which models and modes the subscription exposes. In the API, keep the key on the backend and enforce request limits. Change production dependencies only after access, safety, and useful results are demonstrated in the team’s own environment.
Frequently asked questions
When were the models announced? Meta’s announcement is dated October 12, 2026.
What are the API rates? The documented Standard-mode Muse Wonder rates are $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. Verify the live page before budgeting.
Does Muse Code include Model API credit? Do not assume so. The subscription and API are separate offers with different billing and controls.
Can I send private code? Review the mode, retention, training rules, and organization approval first. Begin with synthetic or approved code.
