Meta’s AI lineup now has a model built for the less glamorous side of AI agents. Muse Ember, the newest release from Meta Superintelligence Labs, is the company’s smallest and cheapest Muse model so far. It’s designed to handle the repetitive tasks agents perform thousands of times a day, rather than impressive one-off reasoning demos.
That focus is deliberate. Meta is positioning Ember for tool calls, coding subtasks, routing requests, extracting data and parsing long documents. In other words, it’s the background work that becomes expensive when running at scale. Pricing starts at 10 cents per million input tokens, which places Ember well below proprietary small models such as Anthropic’s Haiku and OpenAI’s Luna. It also puts Meta in the same price range as low-cost open models like DeepSeek V4.1 Flash and GLM 5.3 Flash. Meta hasn’t highlighted output pricing in the same way, and that matters for agent workloads that generate large amounts of text.
Meta also says Ember is both cheaper and smarter than rival lightweight models, citing Artificial Analysis’s composite intelligence benchmark. Those comparisons come from the company’s own announcement, and composite scores don’t always reflect how a model performs in specific real-world tasks. Independent testing on actual agent workflows will be the more meaningful measure.
Ember’s role becomes clearer when paired with Muse Wonder, Meta’s frontier model released in October. In this setup, the larger model plans the work and sets the rules, while Ember handles the execution. A customer support system, for example, could use Wonder to decide how a certain complaint should be handled and then let Ember apply that approach across tens of thousands of tickets. This planner-and-worker pattern is becoming common across the industry because sending every step to a frontier model gets expensive quickly.
Ember also accepts more than text. Meta trained it to work with images and audio, so agents can process screenshots, charts and scanned documents, or work directly from voice memos, meeting recordings and video without switching to another model. For high-volume automation, avoiding that extra handoff can save both time and money.
The release completes a rapid build-out for the Muse family. Ember follows the mid-sized Spark in April, Glimmer, its open-source counterpart, in August, and Wonder in October. The consumer-focused Muse Agent launched in September and quickly became one of the most popular apps in the US, with features such as Spaces and video chat arriving this month. Meta has steadily covered each part of the AI stack, and Ember now fills the high-volume end. The real test is whether developers trust Meta’s benchmark claims enough to move their large-scale workloads.