Playbooks · Coding tasks
Check whether your session is really on Muse Wonder or Wonder Full — and which route you are paying through
Written October 29, 2026 · Edited November 11, 2026
For a developer or small team: Wonder and Wonder Full are two models sold under one name, through different routes. Muse Wonder is sold per token through the Meta Model API, and Muse Wonder Full comes only with a Muse Code Power or Maximum plan. Assuming a session is on the model or plan it is not can send confidential code to a model labeled "Used to improve our products", or leave a build depending on a model the API does not sell.
Before you start
Meta sells two models under one name, through two different routes. Muse Wonder, model ID muse-wonder-1.0, is available through the Meta Model API and in Muse Code, at $4.00 per million input tokens, $0.20 per million cached input tokens, and $20.00 per million output tokens, with a 1-million-token context window. Muse Wonder Full, model ID muse-wonder-full-1.0, is listed as available by subscription in Muse Code, with no per-token price: it is not sold through the API at any price.
The subscription route is Muse Code, in four plans. Everyday Usage is $5.00 a month and includes Muse Spark, Muse Wonder, and Muse Image, with 10–50 prompts every 5 hours. High Usage is $15.00 a month, at 5 times the Everyday usage. Power Usage is $50.00 a month, at 20 times the Everyday usage, and includes access to the most powerful Muse models, including Muse Wonder Full. Maximum Usage is $200.00 a month, at 80 times the Everyday usage, and includes everything in the Power Usage plan. The lower two plans include Muse Wonder, not Wonder Full.
The second check is the model ID in front of you. On Meta's pricing table, the Muse Spark models come in two forms with the same rates in each form: the -contributor IDs at $0.10 input, $0.002 cached input, and $0.20 output per million tokens, labelled "Used to improve our products," and the plain IDs at $1.25, $0.15, and $4.25, labelled "Not used to improve our products." There is no -contributor form of Muse Wonder on that table. The label on the model ID you are using is the fact to check; the model name alone does not settle it.
What "done" looks like
At the end of one sitting you have, for each repo you work on: the access route you will use, the model ID the session is set to, and a note of which repos may be worked on with a model labelled "Used to improve our products" and which may not. Nothing confidential goes into a session until that note exists.
Steps
- Decide whether you need Wonder Full at all. Wonder Full is reachable only through a Power Usage or Maximum Usage plan, and only inside Muse Code. If the work runs through the Meta Model API — your own code calling the API, a harness, a script — the model available there is Muse Wonder, at the per-token rates above. Do not plan an API build around Wonder Full; the API table lists no price for it because it is not sold there.
- Price one task before choosing the route. At the Wonder rates, 100,000 input tokens and 10,000 output tokens cost $0.60: 100,000 × $4.00 plus 10,000 × $20.00, per million. Compare that with the plan that would cover the same work: Everyday Usage at $5.00 a month covers 10–50 prompts every 5 hours, and the higher plans multiply that allowance. A few sessions a month favour the API route; daily, all-day use favours a plan. The comparison is per-task cost against plan price, not per-token cost against another model's per-token cost.
- Sort your repos before you start a session. Make two lists: repos whose contents you are willing to have used to improve Meta's products, and repos whose contents you are not. The deciding fact is Meta's own label on the model ID, from the step above. A public demo repo and a client codebase can sit in different lists, and the same agent can serve both — one session at a time, with the model ID checked each time.
- Check the model ID at the start of every session. Open the model picker in the session and read the full model ID, including any
-contributorsuffix. If the ID is a-contributorform, Meta labels it "Used to improve our products." If it is the plain form, Meta labels it "Not used to improve our products." If the ID is not the one you intended for this repo, switch it in the picker before typing anything about the repo. Do this check at the start of the session, not after the first prompt: the first prompt is already sent under the model that was selected. - On Power or Maximum, pick Wonder Full deliberately. The plan includes Wonder Full, but inclusion is not selection. When a task is the reason you bought the plan — a long, multi-step build where the stronger model earns its keep — select Muse Wonder Full in the picker and confirm the full model ID shown there matches the one for Wonder Full before starting. When the task is routine, the included Muse Wonder or Muse Spark model at the lower allowance cost is the ordinary choice. Using the strongest model for every task spends the plan's allowance fastest for no gain.
Check the agent's work
- Check that the model ID in the session matches the repo list: no repo from the second list was worked on under a
-contributorID. - Check that no API build depends on Wonder Full. If a script, harness, or integration names Wonder Full, it names a model the API does not sell.
- Check that the plan comparison used the allowance multiples: Power Usage at 20 times Everyday, Maximum Usage at 80 times, not a token count Meta does not publish for plans.
- Check that the per-task price used the Wonder rates: $4.00 input, $0.20 cached input, $20.00 output per million tokens. The Spark rates are different and belong to a different model.
What can go wrong
- Buying a plan for Wonder Full and never selecting it. Power and Maximum include Wonder Full; the session still starts on the model it was set to. Inclusion without selection buys nothing.
- Reading the model name instead of the model ID. "Muse Spark" names two IDs with different labels and very different rates. The suffix is the difference. Read the full ID.
- Checking the model after the first prompt. The check belongs at the start of the session. A prompt sent under the wrong ID cannot be unsent.
- Treating plan allowances as token counts. Plans are described in prompts per 5-hour window and multiples of the Everyday allowance. Meta does not publish a token count for a plan, so a budget built from token counts does not map onto a plan.
Limits
- This page states what Meta's own pages list: plans, model IDs, rates, and the two labels. It does not state how Meta treats content under a subscription beyond those labels, because Meta's pages read for this page do not say more than the labels.
- Wonder Full has no published per-token price and no published benchmark-to-cost figure on the pages read. Any claim about what Wonder Full costs per task would need a plan allowance figure Meta does not publish.
- Prices and plan contents can change. The figures here are the ones on Meta's Muse Code and Model API pages; re-read those pages before paying.
Related playbooks
- FlashProve: proofs of which model answered you
- Check DeepSeek's dsh tool is really running the version with the sandbox-escape fix
- Stay inside the Muse Code subscription allowance
- MOSS-Live: full-duplex spoken AI
Sources
Sources: Muse Code — Meta, Meta Model API — Meta.
Back to: Coding playbooks hub · Also: Token cost calculator