Meta releases Muse Ember at 10 cents per million input tokens on 9 November
Meta Superintelligence Lab released Muse Ember on 9 November, a lightweight model priced at 10 cents per million input tokens for high-volume agent tasks.
Photo: TechCrunchKey points
Meta released Muse Ember, a lightweight AI model priced at 10 cents per million input tokens, designed for high-volume agentic execution.
Meta Superintelligence Lab released Muse Ember on 9 November, its lightest and most cost-efficient AI model yet, priced at 10 cents per million input tokens. The model targets high-volume agentic execution, handling tasks like tool calls, coding subtasks, routing, extraction, and long document parsing, according to Meta's announcement. Meta said Ember is cheaper and smarter than competitors like Anthropic's Haiku and OpenAI's Luna, citing Artificial Analysis's composite benchmark score. At 10 cents per million input tokens, Ember is significantly cheaper than frontier competitors and cost-competitive with open source models like DeepSeek V4.1 Flash and GLM 5.3 Flash.
Pricing and positioning
Ember is designed to pair with a larger model like Meta's Muse Wonder in an agentic harness. In a customer-support scenario, a heavier model establishes rules for handling an issue, then a lighter model applies those rules across tens of thousands of cases. As Meta puts it, "Muse Wonder plans, Muse Ember handles the volume."
Meta trained Ember to work across text, images, and audio, making it flexible in handling inputs. "Agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model," Meta said in the announcement. Ember is the fourth model under Meta's Muse brand, following the mid-weight Spark released in April, Spark's open source twin Glimmer released in August, and the Wonder frontier model released in October. The flagship consumer AI product, the Muse Agent, launched in September with features like Spaces and video chat rolling out earlier this month.
Cloud availability
According to Artificial Analysis benchmarks, Ember reportedly performs on par with models like Anthropic's Claude Haiku 5.5. The model is available through Meta's Model API at $0.10 per million input tokens and $0.50 per million output tokens, as reported. Google Cloud now offers three Muse models—Ember, Spark 1.3, and Wonder—through its managed APIs on the Gemini Enterprise Agent Platform. All three support a 1 million token context window and accept multimodal inputs including text, images, video, audio, and PDFs. The models are also available on Oracle Cloud.
The Muse family marks a shift from Meta's earlier Llama line, offering models through managed APIs as hosted services rather than downloadable weights. This puts Meta's AI inside Google's enterprise storefront, even as Google's Gemini competes directly with Muse. Cloud providers profit from workloads regardless of whose model powers them.
Meta's release coincides with Anthropic's unveiling of Fable 5.2 and Mythos 5.2 on the same day, targeting more demanding technical tasks. Fable can coordinate more than 1,000 AI agents simultaneously and handle hardware-related work. Both companies' announcements highlight a market focused on making AI agents practical for business use. Cost, speed, and the ability to execute complex workflows are becoming as significant as reasoning capabilities. The releases also bring attention to safety, with Anthropic subjecting its new models to longer testing periods and additional evaluations.
Frequently asked questions
What is Muse Ember?
Muse Ember is a lightweight AI model released by Meta Superintelligence Lab on 9 November, designed for high-volume agentic tasks like tool calls, coding subtasks, and document parsing. It costs 10 cents per million input tokens.
How much does Muse Ember cost?
Muse Ember costs $0.10 per million input tokens and $0.50 per million output tokens, making it significantly cheaper than frontier competitors like Claude Haiku 5.5.
Where can I access Muse Ember?
Muse Ember is available through Meta's Model API and on Google Cloud's Gemini Enterprise Agent Platform, as well as Oracle Cloud. It supports a 1 million token context window and multimodal inputs.


