Signal

Muse Ember: Our smallest, most efficient model

First reported by Research.meta ·

The signal ●●●● Compiled by AI from Research.meta, Hacker News, TechCrunch, The Decoder, SiliconANGLE and 7 more
Why you might care

AI agent costs for repetitive tasks may decrease significantly, enabling more complex workflows to run economically.

What happened

Meta AI has launched Muse Ember, a new language model designed for efficiency and complex reasoning tasks such as tool calls, coding subtasks, data extraction, and long document parsing. The company claims Ember offers superior intelligence per dollar compared to competitors like Claude Haiku 5.5 and GPT-6 Luna. Specifically, it matches Haiku's best scores at approximately one-third the cost and outperforms all of Luna's settings on the Artificial Analysis leaderboard for efficiency. Muse Ember is available on the Meta Model API, priced at $0.10 per million input tokens and $0.50 per million output tokens. Partners have already begun integrating Muse Ember into agentic workflows for tasks like contract review and code generation, with endorsements from companies including Box, Cline, Oracle, and Cloudflare highlighting its speed, cost-effectiveness, and capability in handling complex engineering and enterprise content tasks.

What it means

Muse Ember's introduction specifically targets the operational costs associated with running AI agents, which are often burdened by the sheer volume of small, structured tasks they perform. By claiming to offer more intelligence per dollar than leading models for these specific use cases, Meta is positioning Muse Ember as a foundational component for cost-effective agent development and deployment. This could accelerate the adoption of AI agents in business processes by making them more financially viable for companies of all sizes.

The model's design for efficiency in repetitive tasks, coupled with its large context window (1M tokens) and output capacity (up to 512K tokens), suggests a strategic move towards enabling agents to handle more substantial, end-to-end workflows without constant human oversight. This increased capability for autonomous operation on extensive datasets and codebases could redefine how businesses approach automation, particularly in fields involving extensive document analysis or software development.

AI-written summary. May contain errors.

Muse