Meta has introduced Muse Ember via its Superintelligence Lab, a lightweight multi-modal model built for complex reasoning, tool calls, and long-document parsing across text, images, and audio at 10 cents per million input tokens.
The new system was designed for complex thinking, use of tools, programming small pieces of code, managing requests, extracting information, and analyzing large documents.
Low cost and collaboration with larger systems
Meta reported that the cost is ten minutes per million data points in input. According to comparisons using common evaluation criteria, Ember is cheaper and more capable than Anthropic's Haiku and OpenAI's Luna, while being in the same price range as open models such as DeepSeek V4.1 Flash and GLM 5.3 Flash.
The model is intended to work in conjunction with larger Meta systems, such as Muse Wonder.…