AI Directory : AI Agent, AI API, AI Models, Large Language Models (LLMs)
What is Muse Ember?
Muse Ember is the smallest and most cost-efficient model in Meta's Muse family, released on November 9, 2026. Before the launch it ran anonymously on OpenRouter as the stealth model Crouton Alpha, and OpenRouter has since confirmed that Crouton Alpha was an early version of Muse Ember. It is built for the repetitive work inside agent systems: tool calls, coding subtasks, routing, extraction and long document parsing. Developers who were testing Crouton Alpha are the most direct audience, along with teams that need a cheap model to run many agent steps.
How to Use Muse Ember
Meta offers the model through the Meta Model API under the name muse-ember-1.0, and inside Muse Code. The Meta announcement says it supports the Responses API, so an existing agent loop can switch the model name and keep the same tool definitions. On OpenRouter the model is listed as meta/muse-ember; the old stealth listing now points visitors there because the stealth period has ended.
Meta suggests splitting the work: a larger model such as Muse Wonder plans the task, and Muse Ember executes the many bounded steps that follow, such as reading files, calling tools, running tests and making targeted code fixes. Send long inputs in a single call when you need them, since the context window is 1M tokens and output can reach 512K tokens.
Key Features
- 1M-token context Takes in a full codebase or document set in one call, with up to 512K output tokens.
- Multimodal input Accepts text, images, video, audio and PDF documents, and returns text.
- Agent-oriented training Aimed at tool calls in sequence, subagent work, computer use from screenshots and multi-step business workflows.
- Low token price $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens for prompts up to 200K tokens.
- Reasoning effort control OpenRouter lists effort levels from minimal to max for this model.
Why Choose Muse Ember?
Meta positions Muse Ember against Claude Haiku 5.5 and GPT-6 Luna and publishes a side-by-side table. On AutomationBench-AA, Meta reports 37.8% for Muse Ember versus 13.9% for Claude Haiku 5.5 and 20.2% for GPT-6 Luna, while Claude Haiku 5.5 leads on FrontierCode 2.0-AA with 34.9% against Muse Ember's 27.2%. These are Meta's own figures, so test them on your own tasks. Long prompts cost more: above 200K tokens the price rises to $0.30 input and $1.50 output per million tokens.
Use Cases and Applications
Typical jobs include running the test-and-fix loop inside a coding agent, routing incoming requests to the right tool, pulling fields out of long contracts or PDFs, and acting as a subagent under a planner model. Teams that saw Crouton Alpha handle these jobs during the stealth preview can now move the same prompts to a named, documented model.
Frequently Asked Questions
Is Crouton Alpha the same model as Muse Ember?
Yes. OpenRouter states that Crouton Alpha was developed and operated by Meta and was an early version of Muse Ember. The stealth listing is closed and links to the Muse Ember model page.
Is Muse Ember free?
No. Crouton Alpha was free during the stealth preview, but Muse Ember is billed per token: $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 200K tokens.
Were my Crouton Alpha prompts used for training?
OpenRouter says prompts and completions sent to Crouton Alpha may have been retained by the provider but are not used for training, and other use falls under its Stealth Model Terms.
Are the Muse Ember weights open?
Meta's announcement does not offer downloadable weights. Access is through the Meta Model API, Muse Code and OpenRouter.