RC RANDOM CHAOS

Muse Ember Offers Cost-Efficient Reasoning for Agents

· via Hacker News

Original source

Our Smallest, Most Efficient Model: Muse Ember

Hacker News →

Muse Ember is a compact reasoning model designed for high-volume agent tasks including tool invocation, coding subtasks, routing, extraction, and document processing. Testing showed it matches Claude Haiku 5.5 performance at roughly one-third the cost while outperforming GPT-6 Luna across all configurations. Early partnerships demonstrated its utility for contract review and code generation.

The model features a 1-million-token context window, up to 512,000 output tokens, and native multimodal capabilities covering text, images, audio, and video. It is priced at $0.10 per million input tokens and $0.50 per million output tokens on Meta Model API. Safety evaluation encompassed chemical, biological, cybersecurity, and loss-of-control risks through the Meta Superintelligence Scaling Framework, with results remaining below high-risk thresholds. Assessment also covered the model’s behavior when acting on users’ behalf, including intent respect and resistance to embedded instructions.

Muse Ember represents the first model trained using Muse Wonder Full. In the broader model portfolio, it complements Muse Wonder—designated for complex reasoning—and Muse Spark, which handles wider agentic workflows. Operating at approximately $0.14 per Artificial Analysis task, the model is available on Meta Model API, Google Cloud, and Oracle Cloud.

Read the full article

Continue reading at Hacker News →

This is an AI-generated summary. Read the original for the full story.