Meta introduces Muse Wonder with separate access to Full
Meta introduced Muse Wonder and Muse Wonder Full on October 12. Wonder is available through the Meta Model API and Muse Code, while Full is reserved for selected Muse Code subscriptions.
The developer overview explicitly labels Wonder as a preview. Its documented context window holds 1,048,576 tokens. Meta lists text, image, video, audio and PDF inputs, with text output. These specifications describe the service developers can target, rather than a guarantee that every long or complicated task will succeed.
Full access requires a subscription
Meta’s model documentation puts Wonder on every Muse Code plan and on usage based billing. Full is available only on the Power Usage and Maximum Usage subscriptions. Raising Wonder’s reasoning setting to its maximum does not switch to Full, and the documentation lists no API access for Full.
The model catalog advertises Power Usage at $50 a month and Maximum Usage at $200 a month. Full therefore requires a subscription even for developers already paying for Wonder API calls.
Meta plans to bring Wonder to the consumer Muse product, starting with subscribers and supporting artifact generation. The announcement gives no rollout date.
What API usage costs
Published API prices are $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Meta lists no surcharge for filling more of the context window. Web search grounding adds $2.50 per thousand searches on top of token charges.
A Muse Code subscription covers the CLI through its connected subscription credential. Additional API keys remain billed by usage. Plans also have usage limits, so the monthly charge should not be read as an unlimited allowance. Meta says subscription availability and benefits can vary by region.
The data setting matters too
Muse Code subscriptions enable Contributor mode by default. In exchange for more usage, that setting lets Meta use submitted content and generated responses to train future models. Subscribers can turn it off, after which future content is excluded from training unless they enable the setting again.
Wonder API calls use Standard mode, which Meta says excludes submitted content from model training. There is no discounted Contributor mode for Wonder on the API. The subscription’s default therefore deserves a separate check when moving a workflow into Muse Code.
What the capability claims establish
Meta’s research explanation describes Active Memory as retrieving context across conversations and Active Focus as persisting through difficult work. These are intended behaviors, whose reliability requires testing beyond a context window specification.
Meta presents evaluations of expertise, memory, coding and sustained execution. ByteForward has not reproduced these vendor reported tests. Their results cannot establish the success rate a developer would see in a different repository or workflow.
Our earlier coverage of the Muse Code beta explains the terminal agent that preceded this model update.
