Muse Ember: Our smallest, most efficient model
Points and comments are a snapshot, not live.
Meta's Muse Ember offers high intelligence per dollar for agentic tasks.
Meta introduced Muse Ember, its smallest and most efficient model, optimized for complex reasoning, tool calls, coding subtasks, and long document parsing. It claims to deliver more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna, matching Haiku's best score at about a third of the cost. Priced at $0.10 per million input tokens and $0.50 per million output tokens, it features a 1M-token context window and up to 512K output tokens. Muse Ember outperformed competitors on benchmarks like AutomationBench-AA and Agents' Last Exam, and is available on the Meta Model API.
What commenters are saying
The top comment thread featured Simon Willison testing Muse Ember's reasoning levels by prompting it to generate SVG pelicans riding bicycles. Commenters noted a surprising breakdown in the usual correlation between reasoning effort and output quality, with minimal effort sometimes producing better results. Some speculated models may be optimizing for these specific tests. Others discussed the convergence of imagery (fish in baskets, red scarves) as 'ley lines' from training data, and questioned the usefulness of multiple reasoning levels.