Released Muse Ember, a low-cost model aimed at routine agent tasks such as tool use, coding subtasks and reading long documents. (Published November 9.)
Meta Superintelligence Labs released Muse Ember on November 9, its smallest, most cost-efficient model, aimed at the repeated work inside AI agents: tool calls, coding subtasks, routing, extraction and long-document processing. Available through Meta’s API and Muse Code, Ember accepts text, image, audio and video, and offers a 1-million-token context window. For prompts up to 200,000 input tokens, pricing is $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens. Meta reports 37.8% on AutomationBench-AA, versus 13.9% for Claude Haiku 5.5 and 20.2% for GPT-6 Luna; its Terminal-Bench 4.0 score is 40.9%, just above Haiku’s 39.2%. Artificial Analysis estimates about $0.14 per task, against Haiku’s $0.41. The strategy is specialization, not replacing larger models: Meta positions Ember to handle high-volume execution alongside Muse Spark or Muse Wonder, lowering costs for businesses running agents at scale.