Author: Nadali. Prepared by the NADALI editorial team using AI.
Meta introduced Muse Ember on November 9, 2026. The model is available through Meta Model API and Muse Code; the company positions it for repeated work, including ticket classification and document processing. For businesses with long-running work conversations, one launch condition matters: beyond a certain input size, the entire API request moves to higher rates.
Check the threshold
The current pricing documentation lists these rates per million tokens:
- Up to and including 200,000 input tokens: USD 0.10 for input, USD 0.004 for cached input and USD 0.50 for output.
- Above 200,000: USD 0.30, USD 0.012 and USD 1.50, respectively.
Cached input and conversation history continued through previous_response_id count toward the threshold. The higher rates apply to the whole request, not only the excess. Web-search calls cost extra.
A short new question therefore need not mean a short billed input. Ask your developer to show the actual request’s token count, rather than only the latest message.
A numerical example, not a savings claim
Our hypothetical calculation: 199,000 uncached input tokens and 1,000 output tokens cost USD 0.0204; 201,000 uncached input tokens with the same output cost USD 0.0618. This covers token charges at the listed rates only, excludes other costs and is not a result from our own model test.
A useful question for a small-business owner to ask their contractor is whether classifying each new enquiry carries the entire earlier conversation. First establish which information is necessary. Trimming history blindly can remove order conditions or a person’s previous decision.
Check your automation
We suggest a small review using authorised test data.
- Choose a short and a long version of the same task, without real contact details or secrets.
- Record the complete input, output, cached-token count and applied pricing tier.
- Check answer accuracy, retries and the cost of human review.
- Define a budget and what happens as it approaches its limit before processing customer work at scale.
This is editorial advice, not an integration procedure we have tested. The test assistant should not be able to change orders or send customer replies on its own.
API usage and subscriptions differ
The Muse Ember reference says the model runs in Standard mode only on Model API, without a Contributor variant. In Muse Code it is available on every plan; a request above 200,000 input tokens consumes subscription usage faster. The calculation above concerns API charges, not a monthly subscription.
We have not tested access from a Ukrainian account, a specific integration or output quality. Check availability in your service and its contractual data-handling terms before switching. A low token rate alone does not establish that an accepted result will cost less.
Sources checked November 11, 2026. Cover: a conceptual AI illustration, not Meta’s interface.
