Meta releases Muse Ember, its lightest model, at 10 cents per million input tokens

Meta Superintelligence Lab released Muse Ember on Monday, the lab's lightest and most cost-efficient AI model to date, priced at 10 cents per million input tokens. TechCrunch reported the release.

Meta says Ember is "designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing."

Pricing and cost claims

At 10 cents per million input tokens, Ember is significantly cheaper than frontier competitors such as Anthropic's Haiku and OpenAI's Luna, according to the report. It is also cost-competitive with open-source models including DeepSeek V4.1 Flash and GLM 5.3 Flash.

Meta emphasized cost-per-task in its release. It uses Artificial Analysis's composite benchmark score to claim that Ember is both cheaper and smarter than Haiku and Luna. The specific score figures were not given in the report. Output token pricing, context window and parameter count are not yet disclosed.

Designed to pair with Muse Wonder

Meta positions Ember as the high-volume worker in an agentic harness alongside a larger model, Muse Wonder. In a customer-support example, a heavier model sets the rules for handling an issue, then a lighter model applies those rules across tens of thousands of cases. Meta summarizes it this way: "Muse Wonder plans, Muse Ember handles the volume."

Multimodal input

Meta trained Ember to work across text, images and audio. In the announcement, Meta says "agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model."

Place in the Muse lineup

Ember is the fourth model under Meta's Muse brand:

• Spark: mid-weight model, released in April.
• Glimmer: Spark's open-source twin, released in August.
• Wonder: frontier model, released in October.
• Ember: lightest model, released Monday.

Separately, the consumer product Muse Agent was released in September. Features such as Spaces and video chat began rolling out earlier this month, according to TechCrunch.

What this means

Meta is selling Ember on cost-per-task rather than peak capability. Its framing is a two-tier system: Wonder handles planning and Ember executes repetitive work at volume. The 10 cents per million input tokens price targets workloads such as routing, extraction and document parsing, where call volume dominates cost.

Native handling of text, images and audio means agent pipelines can process screenshots, scanned documents and recordings with one model instead of adding separate ones. The cost and intelligence comparisons against Haiku, Luna, DeepSeek V4.1 Flash and GLM 5.3 Flash rest on Meta's use of the Artificial Analysis composite score. Teams evaluating Ember should test it on their own task mix, particularly because output pricing and context window have not yet been disclosed.