Meta Releases Muse Ember, a Lightweight AI Model Built for High-Volume Agent Workloads
Meta Superintelligence Lab has launched Muse Ember, a low-cost, lightweight AI model engineered for repetitive agent tasks and multimodal execution at 10 cents per million input tokens.
Meta Superintelligence Lab introduced a new artificial intelligence model on Monday named Muse Ember, expanding its Muse product line with a focused emphasis on efficient agent execution. Positioned as Meta's lightest and most cost-effective model to date, Ember is engineered specifically for complex reasoning execution alongside the repetitive subtasks that automated agents perform continuously. Key targeted workloads include handling tool calls, executing code subtasks, routing data, extracting information, and parsing lengthy documents.
Pricing performance and multimodal capabilities
Cost efficiency is a primary structural element of Ember. Meta set the model's pricing at 10 cents per million input tokens to accommodate heavy execution requirements. In its announcement materials, Meta pointed to Artificial Analysis’s composite benchmark score to demonstrate that Ember delivers higher accuracy at a lower cost than competing lightweight proprietary models, specifically Anthropic’s Haiku and OpenAI’s Luna. Additionally, Meta noted that Ember is cost-competitive with open-source Flash models such as DeepSeek V4.1 Flash and GLM 5.3 Flash.
Beyond low token pricing, Ember features native support for multimodal inputs across text, image, and audio formats. Meta trained the model to handle diverse media types directly within agent workflows. As Meta described in the release, agents operating on Ember can analyze screenshots, charts, and scanned documents, or process audio formats including voice memos, meeting recordings, and video feeds without relying on auxiliary specialized models.
How Ember pairs with frontier models
Ember is structured to operate within an agentic harness alongside larger, more capable foundation models. Instead of functioning as a standalone general-purpose solution, it is designed to complement heavy frontier systems such as Meta’s Muse Wonder. Meta described this operational split in the announcement with the statement: “Muse Wonder plans, Muse Ember handles the volume.”
This workflow structure allows developers to assign high-level strategy and rule creation to a larger model while using Ember to execute those rules across high-volume tasks. In a customer-support system, for instance, a heavier model like Muse Wonder can formulate the logic required to address specific inquiries, while Ember applies those established rules across tens of thousands of individual user tickets to manage operational costs.
What it means for developers
The introduction of targeted execution models like Ember highlights a shift toward multi-tier agent architectures. Rather than using expensive flagship models for every step of a software process, developers can separate high-level reasoning from high-frequency downstream execution to reduce operational expenditure.
With native multimodal support and low per-token costs, models like Ember allow teams to streamline multi-step pipelines without integrating separate external converters for images or audio assets. When building and testing multi-tiered architectures, developers can try top AI models cheaply through one API at https://apixoai.online to compare how different model combinations perform under practical workloads.
Ember represents the fourth AI model released under Meta's Muse brand. It follows the mid-weight Spark model released in April, Spark’s open-source counterpart Glimmer in August, and the Wonder frontier model in October. These releases accompany Meta's consumer product, the Muse Agent, which launched in September and recently introduced features such as Spaces and video chat.
Source: Meta releases Muse Ember, a new lightweight model for agent execution — TechCrunch AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
