Skip to content
MaungaAI news, explained

Models

Meta says its smallest AI does the same work for a third of the cost

For short texts, Muse Ember charges the same as its rivals from Anthropic and OpenAI. According to Meta, the difference is that it uses much less to finish each task. The catch is that it only runs on the company's servers.

2 min read3 sourcesTranslated with AI from the Spanish original

Share: WhatsApp · Email

AI-generated illustration for Maunga. Not a photograph.

The essentials

  • Meta launched Muse Ember, its smallest model, on November 9. It is designed for the repetitive tasks of agents, AI programs that work on their own (TechCrunch, MarkTechPost).
  • For short texts, it charges the same as Claude Haiku 5.5 and GPT-6 Luna: $0.10 per million tokens (chunks of text) it receives (MarkTechPost, Simon Willison's Weblog).
  • Claimed by one partyAccording to Meta, completing each task in the Artificial Analysis index with Ember costs $0.136, compared with $0.411 with Claude Haiku 5.5 (MarkTechPost).
  • Claimed by one partyIt can only be used on Meta's servers, through Meta Model API and Muse Code. Its weights are not public, according to MarkTechPost (MarkTechPost).

Why it matters

Price per word is no longer enough to compare these AIs. For short texts, the three models charge the same. If Meta's figures are confirmed, Ember would need less text to solve each task. That would make it cheaper for companies to run agents that repeat the same work thousands of times, such as serving customers or reading documents.

The details

According to Meta's figures, as reported by MarkTechPost, Ember ties with Claude Haiku 5.5 at 36 points on the Artificial Analysis index, both at their highest reasoning level. That index is an average score across several tests. Ember stands out most in automated office workflows, completing 37.8% of tasks. It falls behind Haiku on a programming test that uses real code (MarkTechPost).

It understands text, images, audio and video without help from another model (TechCrunch, MarkTechPost). Meta presents it as a partner to Muse Wonder: the large model plans and Ember handles the volume (TechCrunch). For long texts, the sources disagree. Simon Willison says Ember's price rises to $0.30 per million above 200,000 tokens (Simon Willison's Weblog). MarkTechPost says Meta does not publish a higher rate for them (MarkTechPost).

What we don't know

  • Whether the savings per task hold up in real use, outside the tests Meta publishes.
  • How much it costs with very long texts: Simon Willison and MarkTechPost give different accounts.

Sources

This story is based on 3 sources. In the text, each fact names in brackets who reports it.

Was this story useful? Yes Not really

Spanish original