Meta launches Muse Ember for agent scale

Muse Ember is a strategically sharp release: agent systems need inexpensive worker models more than another expensive general-purpose flagship.

• –Its 1M-token context window supports full codebases, document sets, and multimodal inputs in a single workflow.
• –Meta positions it as a subagent model that handles repeated execution while Muse Spark or Muse Wonder manages planning.
• –Early results show strong agentic and coding performance, though most comparisons rely on Meta or partner evaluations.
• –The low pricing makes high-volume routing, extraction, and tool-use workloads far more economically viable.
• –Developers can access it through an OpenAI-compatible Responses API, reducing migration friction.