One way to make a cheap model expensive is to send it the entire job every time.
Meta has added Muse Ember to its Model API and Muse Code. The developer announcement describes a small model for repeated work such as tool calls, document extraction, and coding subtasks. Meta says Ember is available across Muse Code plans. The API’s token rates are separate from the terms of those subscriptions.
Watch the prompt that crosses the threshold
Meta lists $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 200,000 input tokens. Above that threshold, the listed rates are $0.30 and $1.50. A growing tool transcript can therefore affect the cost of the next call, even if the model returns a short answer.
Measure how often a step needs to go back to a larger model
Start with one bounded task such as extracting the fields needed to open a support case. Give Ember the relevant document and a small schema. In a second run, pass the surrounding transcript too. Check how often each approach omits a required field or needs to hand the task to a stronger model.
Count the follow-up calls when you calculate the cost. A cheaper first call can lose its advantage if the next model has to redo the work. We have not measured Ember’s handoff rate. For a separate Muse workflow, see our Muse Code handoff guide.
Abdessalam Alaoui
I'm Abdessalam Alaoui, founder of Musthave.AI. I write practical guides to AI models, tools, generative media.
