TL;DR: We just launched Muse Ember – our most efficient model, and the cheapest way to run frontier-quality agents at scale.
Meta launched Muse Ember, its most efficient model, offering frontier-quality AI agents at the lowest price.Summary
Prashant Ratanchandani announces the launch of Muse Ember, described as the most efficient model and the cheapest way to run frontier-quality agents at scale. Available on the Meta Model API and Muse Code, Ember is purpose-built for agents that spend most tokens on small, structured, repeated steps such as tool calls, coding subtasks, routing, extraction, and long-document parsing. Priced at $0.10 per 1M input tokens and $0.50 per 1M output tokens, with cached input at $0.004, it costs about $0.14 per AA task, placing it beyond the pareto frontier. It features a 1M context window. Ratanchandani emphasizes that frontier capability only matters if the price allows it to be run everywhere for everyone, and Ember delivers Spark-class intelligence at 10% of the price. It is available at dev.meta.ai and through partners including OpenCode, OpenRouter, and Cline.
(Source：LinkedIn)