Meta launches Muse Ember, its cheapest AI model yet

Meta Superintelligence Lab released Muse Ember on Monday, calling it the lab’s lightest and most cost-efficient model to date, built for the repetitive work AI agents handle constantly — tool calls, coding subtasks, routing, data extraction and parsing long documents.

Meta framed the pitch around cost per task, pointing to a composite benchmark score from Artificial Analysis that puts Ember ahead of rivals such as Anthropic’s Haiku and OpenAI’s Luna on both price and capability. Priced at 10 cents per million input tokens, Ember comes in well below those frontier competitors and lands roughly on par on cost with open-source rivals DeepSeek V4.1 Flash and GLM 5.3 Flash.

Meta designed Ember to work alongside a larger model such as its own Muse Wonder inside an agentic harness: the heavier model sets the rules for handling a task, while the lighter one applies them at scale, such as across tens of thousands of customer-support cases. “Muse Wonder plans, Muse Ember handles the volume,” Meta said in its announcement. Meta also built Ember to handle multiple input types at once — text, images and audio — so a single model can process a screenshot, a chart, a scanned document, a voice memo or a video clip instead of routing each format to a different tool.

Ember is the fourth model released under Meta’s Muse brand, following the mid-weight Spark in April, Spark’s open-source twin Glimmer in August, and the frontier-class Wonder in October. The brand’s flagship consumer product, Muse Agent, launched in September to unexpected success, with new features including Spaces and video chat rolling out earlier this month.

Why it matters

Ember’s price puts direct pressure on Anthropic’s Haiku and OpenAI’s Luna in the cheap-inference tier that increasingly sets the real cost of running AI agents at scale, where sheer volume — not peak capability — determines the bill.