Meta launches Muse Ember, a cheap, multi‑modal model for AI agents

Meta’s latest AI model, Muse Ember, is now live and promises to make agent‑driven automation cheaper and more flexible. At 10 cents per million input tokens, the model is positioned as the most cost‑efficient option for routine tool calls, coding subtasks, and long‑document parsing.

What happened

According to TechCrunch, Meta’s Superintelligence Lab unveiled Muse Ember on Monday as the lightest and most cost‑efficient model in its Muse lineup. The announcement frames Ember as a workhorse for “complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing.”

The company highlighted a benchmark from Artificial Analysis, claiming Ember is both cheaper and smarter than rivals such as Anthropic’s Haiku and OpenAI’s Luna. At a price of 10 cents per million input tokens, Ember undercuts those frontier competitors and even matches open‑source offerings like DeepSeek V4.1 Flash and GLM 5.3 Flash on cost.

Ember is not a stand‑alone chatbot; Meta positions it as the “volume” half of an agentic harness paired with a larger model, Muse Wonder. In a customer‑support scenario, Wonder would set the policy while Ember applies it across tens of thousands of tickets—Meta sums it up as “Muse Wonder plans, Muse Ember handles the volume.”

Beyond pricing, Ember is trained to handle multiple media types. The model can ingest text, images, and audio, allowing agents to read screenshots, charts, scanned documents, voice memos, meeting recordings, and video without swapping models. This flexibility is meant to simplify pipelines that previously required separate specialized models.

Ember is the fourth release under the Muse brand, following Spark (April), its open‑source twin Glimmer (August), and the frontier Wonder model (October). The flagship consumer product, Muse Agent, debuted in September and recently added features like Spaces and video chat.

Why it matters for small‑business leaders

For owners and team leads running lean operations, the cost of AI‑driven automation can be a make‑or‑break factor. At 10 cents per million tokens, Ember translates to roughly $0.001 per 1,000 characters of input—orders of magnitude lower than many proprietary agents. That price point makes it feasible to run high‑volume tasks—such as bulk ticket triage, invoice extraction, or media‑rich content moderation—without blowing the budget.

The multi‑modal capability also reduces the engineering overhead of stitching together separate OCR, speech‑to‑text, and image‑analysis services. A single Ember‑powered agent can read a scanned receipt, pull the amount, and log it into an accounting system, all in one pass. Small teams can therefore build end‑to‑end workflows faster and with fewer vendor contracts.

Finally, the “two‑model” strategy (Wonder + Ember) offers a pragmatic scaling path: use a heavyweight model for strategic reasoning and a lightweight Ember for repetitive execution. This mirrors the way many SaaS tools separate orchestration from worker nodes, but now it can be done in‑house with a single vendor stack, simplifying compliance and data governance.

What to do this week

1. Sign up for Meta’s API access and run a quick token‑cost test on a sample document to verify the 10 ¢/M‑token claim for your typical workload.
2. Prototype a simple agent that uses Ember to extract text from screenshots or PDFs and feed the output into your existing CRM or accounting tool.
3. Map any high‑volume, rule‑based processes in your business and draft a two‑model plan: designate Muse Wonder (or another LLM you already use) for policy creation and Ember for bulk execution.

Sources

• Meta releases Muse Ember, a new lightweight model for agent execution — TechCrunch - AI