Meta releases Muse Ember, a low-cost lightweight model built for AI agent workloads

Meta Superintelligence Labs released a new model called Muse Ember on Monday, November 9, the lab's smallest and most cost-efficient model to date. Meta said it was "designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing." The launch was reported by TechCrunch and The Verge.

Price and performance

Muse Ember costs $0.10 per million input tokens and $0.50 per million output tokens. Meta claims it delivers more intelligence per dollar than Anthropic's Claude Haiku 5.5 or OpenAI's GPT-6 Luna at any reasoning setting, matching Haiku's best score at about a third of the cost. TechCrunch noted it is also cost-competitive with open-source models such as DeepSeek V4.1 Flash and GLM 5.3 Flash. The comparisons are based on Meta's own and Artificial Analysis benchmark results.

"Muse Wonder plans, Muse Ember handles the volume"

Ember is designed to pair with Meta's larger Muse Wonder model in an agentic setup. In a customer-support agent, for example, a heavier model might set rules for handling an issue, then hand off to the lighter model to apply them across tens of thousands of cases.

The model natively understands text, images, audio and video, so agents can "read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model," Meta said. It has a 1-million-token context window and up to 512,000 output tokens.

Safety evaluation

Meta said it evaluated Muse Ember under its Superintelligence Scaling Framework, covering chemical and biological, cybersecurity and loss-of-control risks, and found it remained below the framework's high-risk thresholds.

Ember is the fourth model under the Muse brand, following Spark (April), its open-source twin Glimmer (August) and the Wonder frontier model (October). It is available as muse-ember-1.0 on the Meta Model API and in Muse Code.

Sources: TechCrunch, Meta, The Verge