Meta Releases Muse Ember, a Low-Cost Lightweight AI Model for Agents

What Happened

Meta Superintelligence Lab released a new AI model called Muse Ember on Monday, 9 November 2026, according to TechCrunch. Meta describes it as the lab’s lightest and most cost-efficient AI model yet [1]. The Verge describes the launch as Meta’s “smallest and most efficient” AI model [2]. The two reports use different stage words, with TechCrunch saying Meta released the model and The Verge saying it is launching it [1][2].

Meta said Ember is designed for complex reasoning and for the repetitive work that AI agents do all day. It named several examples: tool calls, coding subtasks, routing, extraction and long document parsing [1][2].

• Price: 10 cents per million input tokens [1].
• Meta says Ember is significantly cheaper than frontier competitors Anthropic’s Haiku and OpenAI’s Luna [1].
• Meta says Ember is cost-competitive with open source models DeepSeek V4.1 Flash and GLM 5.3 Flash [1].
• The Verge reports Meta claims Ember gives “more intelligence per dollar” than Claude Haiku 5.5 or GPT-6 Luna “at any reasoning setting” [2].

Meta’s cost comparisons rely on Artificial Analysis’s composite benchmark score, which Meta used to show Ember as both cheaper and smarter than competitors on cost-per-task [1]. These are Meta’s claims as reported by the two outlets.

Meta says Ember is designed to work alongside a larger model, Muse Wonder, inside an agentic harness. TechCrunch gives a customer-support example: the heavier model sets the rules for handling a particular issue, and the lighter model applies those rules across tens of thousands of cases [1]. Meta puts it as Wonder planning and Ember handling the volume [1].

Meta also trained Ember on several types of media, including text, images and audio. According to the announcement quoted by TechCrunch, agents can read screenshots, charts and scanned documents, or work from voice memos, meeting recordings and video, without a separate model [1].

Ember is the fourth model released under the Muse brand. The mid-weight Spark was released in April, its open source twin Glimmer in August, and the Wonder frontier model in October [1]. TechCrunch also reports that the consumer Muse Agent product came out in September and that features such as Spaces and video chat began rolling out earlier in the month [1].

Who Is Affected

TechCrunch names developers building AI agents as the users Meta is targeting with Ember’s low cost and task focus, including those running tool calls, coding subtasks and document parsing [1][2].