Skip to content
AI news · AI tools

Meta's Muse Ember is a budget AI model for the dull, repetitive jobs, priced like the house wine

By , Founder of Adaptd ·

Meta released Muse Ember on 9 November, a small, cheap model built for repeat jobs like sorting, routing and pulling details out of documents. It costs US$0.10 per million input tokens.

The short version

  • Muse Ember is Meta's smallest and cheapest AI model, built for repetitive automation work like routing, extraction and reading long documents.
  • It's available through the Meta Model API and Muse Code at US$0.10 per million input tokens and US$0.50 per million output tokens.
  • If you pay for an automation that runs thousands of small AI steps, ask whoever built it whether a cheaper model would do.

Not every AI job needs a genius. Some just need someone reliable who'll sort the mail all day without asking for a raise.

What is Muse Ember?

Muse Ember is a small, low-cost AI model from Meta, announced on 9 November 2026. Meta built it for repetitive agent work: calling tools, routing requests, pulling details out of text and reading long documents.

Meta says Ember scored 37.8% on an automation benchmark where Claude Haiku 5.5 scored 13.9%, and that it matches Haiku's best result at about a third of the cost. Those are Meta's own numbers, so treat them as a sales pitch until independent testers weigh in.

What does it cost, and where can you get it?

Muse Ember costs US$0.10 per million input tokens and US$0.50 per million output tokens through the Meta Model API and Muse Code. It can take in about a million tokens at once, which is a very long document.

Meta's announcement doesn't mention the Meta AI app or which countries can use the API, so check availability before anyone plans around it.

Should your business use it?

Most small businesses won't use Muse Ember directly, because it's a developer model rather than a chat app. It matters if you pay for automations that run lots of small AI steps, like sorting enquiries or reading invoices.

My advice: ask whoever built your automations which model each step uses and what it costs each month. Plenty of workflows run every step on a premium model out of habit. A cheap model for the sorting and a smart one for the tricky bits can cut the bill, but test it on a week of real data first.

You wouldn't pay a barrister to sort the post. Your automations shouldn't either.

Wondering what this means for your business?

That's exactly what the discovery call is for.