Introducing Our Smallest, Most Efficient Model: Muse Ember

research.meta.ai4 min read

Meta has introduced Muse Ember, a small and efficient AI model designed for complex reasoning execution and repetitive agent tasks such as tool calls, coding subtasks, routing, extraction, and long document parsing. Available today on the Meta Model API and Muse Code, the model costs $0.10 per million input tokens and $0.50 per million output tokens while supporting a 1M-token context window and up to 512K output tokens. Evaluated under the Meta Superintelligence Scaling Framework, the company states that Muse Ember remained below high-risk thresholds while delivering intelligence and efficiency tailored for high-throughput agentic workloads. Open article →

Notable comments

Reception focused on Muse Ember’s claimed strong cost/performance for agentic and long-context tasks, alongside skepticism about benchmark validity/comparability and references to the model’s prior “stealth” identity via routing platforms.

HN · ProblemExplorer Claims large benchmark cost advantages for Ember over other frontier/similar models (using Artificial Analysis/Vals finance-legal agent results) and argues it lands in a similar intelligence range at a fraction of the measured per-task cost.

HN · shlewis Provides an update tying “Crouton Alpha” to an earlier Muse Ember version and clarifies OpenRouter/stealth-model handling and terms, pointing readers to the correct public endpoint post-stealth.

More on this topic