Meta Launches Muse Ember: A Smaller, Cost Effective AI Model for Businesses
Meta has unveiled Muse Ember, a smaller AI model designed for large scale, repetitive tasks managed by AI agents. Announced on November 9, 2026, the model targets businesses and developers looking to automate coding, document processing, data extraction, and other routine workflows at a lower cost. CEO Mark Zuckerberg highlighted that Ember offers more intelligence per dollar compared to competitors such as Anthropic’s Claude Haiku 5.5 and OpenAI’s GPT-6 Luna.
Museumember is part of Meta’s Muse family, complementing larger models rather than replacing them. Meta explains that while bigger models handle complex planning, Ember excels in repetitive execution. Priced at $0.10 per million input tokens and $0.50 per million output tokens, Ember is positioned as a cost effective solution.
Muse Ember, the smallest model in the Muse family, is designed for agentic workflows, executing multiple smaller tasks such as coding subtasks, making tool calls, extracting information, routing, and analyzing long documents. It is trained using Muse WonderFull, a novel model developed by Meta, according to the company.
Ember supports various formats, including text, images, audio, and video, with a context window of 1 million tokens and up to 512,000 output tokens. This enables it to handle big documents, screenshots, charts, meeting recordings, and videos without needing multiple models.
According to Meta, Muse Wonder plans the tasks, while Ember handles the execution in volume. Benchmarks show that Ember scores 40.9% on Terminal Bench 4.0, outperforming Claude Haiku 5.5 (39.2%) and GPT-6 Luna (16.4%). Ember also topped the Artificial Intelligence Index with a score of 36, matching Claude Haiku 5.5 and outperforming GPT-6 Luna (28).
However, Ember did not lead in every benchmark. Claude Haiku 5.5 scored higher on FrontierCode 2.0-AA, recording 34.9% against Ember’s 27.2%.
Meta emphasized its cost efficiency, estimating that an Ember model costs $0.14 per artificial analysis task, compared to $2.39 per task for its own Muse Spark 1.0 model. Muse Ember is available through the Meta Model API, with pricing details as follows: $0.10 per million input tokens, $0.004 per million cached input tokens, and $0.50 per million output tokens.
Google Cloud and Oracle Cloud now offer Muse Ember, Muse Spark, and Muse Wonder, indicating Meta’s intention to make advanced AI capabilities more accessible at scale.
Thinking about building an AI product?
Get in Touch