Meta releases multimodal Muse Ember model for low-cost agentic tasks
Meta has released the Muse Ember model, the cheapest model in the Muse lineup (10 cents per million input tokens), designed for high-volume agentic tasks such as tool calls, data extraction, and document parsing. It supports text, images, and audio, and is designed to be paired with the Muse Wonder model.
On Monday, Meta released the Muse Ember model, which the company says is the lightest and cheapest model under the Muse brand. According to the company, it is designed for "complex reasoning execution and work that agents repeat all day long": tool calls, coding subtasks, routing, data extraction, and parsing long documents. The price is 10 cents per million input tokens.
In comparison with the competition, Meta refers to the composite score of the Artificial Analysis benchmark, according to which Ember is cheaper and more capable than Anthropic's Claude Haiku model and OpenAI's Luna model. According to the company, it is also price-competitive with the open-source models DeepSeek V4.1 Flash and GLM 5.3 Flash.
Ember is designed to be paired with the larger Muse Wonder model in an agentic framework — according to the company, "Muse Wonder plans, Muse Ember handles the volume." The model is multimodal and, according to the company, can process screenshots, charts, and scanned documents, or work from voice recordings, meeting recordings, and video without the need for a separate model.
Ember is the fourth model under the Muse brand, following Spark (April), its open-source variant Glimmer (August), and the Wonder model (October). The flagship consumer product Muse Agent was released in September.
Why it matters
According to Meta, the model offers a low cost per token while retaining multimodal capabilities, which is key for running agents at high volume, where the same task is repeated thousands of times. For companies, this means a potential reduction in inference costs for automated agentic processes; for developers, it adds another low-cost alternative to choose from among models such as Claude Haiku, Luna, or the open-source GLM and DeepSeek.
Release card
Muse Ember
Meta
- Inputs
- text, images, and audio as input; according to the source it can also handle screenshots, charts, scanned documents, voice recordings, meeting recordings, and video
- Price
- 10 cents per million input tokens
- high-volume agentic tasks such as tool calls, routing, and coding subtasks
- data extraction and parsing of long documents
- processing screenshots, charts, voice recordings, and video without a separate model
According to Meta, the model is cheaper and more capable than Anthropic's Claude Haiku model and OpenAI's Luna model based on the Artificial Analysis benchmark's composite score, and is also price-competitive with the open-source models DeepSeek V4.1 Flash and GLM 5.3 Flash.
The card summarizes information from the article and any dated corrections, with a link to the original source. It is not our assessment of the model. It does not yet have a dedicated editorial profile. Model selection and other announcements →
Two audiences, two different impacts
What this means
For individuals
Developers building AI agents get another low-cost multimodal option for high-volume tasks, which can be compared with models such as Claude Haiku or GLM Flash.
For a business
Companies running agentic automation at high volume (e.g. customer support) get, according to Meta, a cheaper option for repetitive tasks such as tool calls, data extraction, and document processing, when combined with the planning model Muse Wonder.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.