Meta Muse Ember
Meta subagent/executor model: 36 on the Artificial Analysis Intelligence Index (tied with Claude Haiku 5.5 max) at ~$0.14 per task, 37.8 AutomationBench-AA, 40.9 Terminal-Bench 4.0, 69.5 OSWorld 2.1; trails Haiku on FrontierCode 2.0-AA (27.2 vs 34.9).
Interactive Token Cost Estimator (Muse Ember)
Architectural Strengths & Target Workloads
- Very low cost per task and $0.004 cached input
- 1M context with 512K max output
- Native vision, audio and video input for document and screen work
Primary Engineering Workloads: Explorer / Reader Subagents · Long-Document Extraction · High-Volume Routing & Tool Calls.
Frequently Asked Questions (Muse Ember)
How much does Meta Muse Ember cost per 1M tokens?
Meta Muse Ember is priced at $0.10 per 1M uncached input tokens, $0.00 per 1M cached input tokens, and $0.50 per 1M output tokens (muse-ember-1.0 on Meta Model API (Responses API); prompts over 200K tokens cost $0.30 / $0.012 cached / $1.50).
What is the context window of Meta Muse Ember?
It supports up to 1M tokens (1,000,000 tokens) and a maximum generation output of 512,000 tokens.
Meta Muse Ember: A $0.10-per-Million Subagent Model With a 1M-Token Context
Meta Superintelligence Labs' smallest model matches Claude Haiku 5.5 on the Artificial Analysis index at about a third of the cost per task. It is built to do the volume work while a bigger model plans.
Read Full Technical Dispatch (5 min read) →