Releasing Muse Ember today — our smallest model that delivers higher intelligence per dollar than others in its class. Built for token-heavy work that agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing. Available today on the Meta Model API and in Muse Code.
Meta CEO Mark Zuckerberg announced Muse Ember on X on 9 November. The post describes it as Meta‘s smallest model and lists the agent work it is built for: tool calls, coding subtasks, routing, extraction and long document parsing. The intelligence-per-dollar line is Meta’s claim; the piece sets it against the vendors’ own pricing pages.
$META JUST LAUNCHED MUSE EMBER, ITS SMALLEST AI MODEL
Meta says Muse Ember is its smallest and most efficient model, built for complex reasoning.
It’s available today on Meta’s Model API for $0.10 per million input tokens and $0.50 per million output tokens. Tokens are the small chunks of text an AI model reads and writes.
That’s about 12x cheaper on input than Muse Spark 1.3, which lists at $1.25, and 40x cheaper than Muse Wonder, Meta’s priciest model at $4.
It comes as Meta builds a sales team to sell its AI to businesses, working with consulting firms like Accenture and Deloitte, according to The Information.
Wall St Engine flagged the launch minutes later with Meta’s input prices for Muse Ember, Muse Spark 1.3 and Muse Wonder. The piece checks those prices on Meta’s own model pages; the input gap to Spark 1.3 works out to 12.5x, which the post rounds to about 12x.
Meta launched Muse Ember on 9 November, which it calls its smallest and most efficient model, at $0.10 per million input tokens and $0.50 per million output tokens on the Meta Model API, according to its research blog. Meta’s own model page lists Muse Wonder at $4.00 and $20.00. That makes Wonder 40x the price of Ember on both input and output, by our arithmetic.
Mark Zuckerberg posted on X: “Built for token-heavy work that agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing”. The same $0.10 and $0.50 list price appears on Anthropic’s page for Claude Haiku 5.5 and OpenAI’s page for GPT-6 Luna, so Meta makes its case on cost per task instead.
Business Pill · Usage-Based Pricing
An evergreen Business Engineer explainer on usage-based pricing. It teaches the general idea only and does not discuss Meta or Muse Ember.
The key insight: Meta priced Muse Ember at the same $0.10 and $0.50 list price that Anthropic and OpenAI show for Claude Haiku 5.5 and GPT-6 Luna, and 40x below its own Muse Wonder by our arithmetic, so it competes on cost per task: in Meta’s own example Ember read 97% of the input tokens for 46% of the bill, by our arithmetic.
What Meta Launched
Meta’s research blog says Muse Ember is “our smallest and most efficient model designed for complex reasoning execution and the work agents repeat all day”. It is available on Muse Code and the Meta Model API as muse-ember-1.0, with a 1M-token context window and up to 512K output tokens, and takes text, image, audio and video inputs, according to Meta’s developer blog.
Meta says Muse Ember is the first model trained using Muse Wonder Full. Before release, it was evaluated under Meta’s Superintelligence Scaling Framework, and Meta says it “remained below the framework’s high-risk thresholds”.
List price per million output tokens (input price in the label), standard API tier. Sources: Meta Muse model pages on dev.meta.ai; Anthropic, OpenAI and Google API pricing pages, all read 9 November 2026. The 40x, 12.5x and 8.5x ratios are our arithmetic.
The 40x Price Ladder
Meta’s Muse Ember page lists $0.10 input, $0.004 cached input and $0.50 output per million tokens for prompts under 200,000 tokens, rising to $0.30 and $1.50 above that. The Muse Spark page lists Muse Spark 1.3 at $1.25 input and $4.25 output. That puts Spark 1.3 at 12.5x Ember on input and 8.5x on output, by our arithmetic. The Wall St Engine post that flagged the launch rounded the input gap to about 12x.
The Spark page also lists a muse-spark-1.3-contributor tier, marked “Used to improve our products.”, at $0.10 input and $0.20 output. Ember’s input price matches that tier; its output price is higher. Meta’s blog names the roles: “Muse Wonder for the hardest problems, Muse Spark for coding and agentic workflows, and Muse Ember for repeatable coding and agentic tasks at scale”.
Priced for the Agent Loop
Meta’s developer blog tells builders to “Use Muse Ember for repeated, bounded work and pair it with Muse Spark or Muse Wonder when a workflow needs deeper planning and reasoning”. Its worked example ported 16 Python lexers to Rust, with Muse Wonder planning and 16 Ember agents doing the ports.
Meta reports that “Muse Ember handled 158M input tokens to Muse Wonder’s 4.4M”, and that the run cost about $3.96: $2.12 for Wonder and $1.84 for Ember. By our arithmetic, Ember read 97% of the input tokens and took 46% of the bill. Meta adds that “Results vary from run to run”.
Cline founder Saoud Rizwan, quoted by Meta, said the model suits “teams cutting costs by running it as the subagent in their harness”.
Same List Price as Haiku and Luna
On Anthropic’s pricing page, Claude Haiku 5.5 costs $0.10 input and $0.50 output for prompts up to 100,000 tokens, and $0.50 and $2.50 above that. OpenAI’s pricing page lists gpt-6-luna at $0.10 and $0.50. Google’s pricing page lists Gemini 3.5 Flash-Lite at $0.30 and $2.50, which is 3x Ember on input and 5x on output, by our arithmetic.
Meta’s claim rests on Artificial Analysis cost-per-task results. Its blog says Muse Ember “matches Haiku’s best score at about a third of the cost”: $0.136 per task for Ember at max effort against $0.411 for Haiku 5.5, in Meta’s chart. The same blog notes “Muse Ember costs are estimates at proposed pricing”.
Meta’s own table is not a clean sweep. Claude Haiku 5.5 scores 34.9 on FrontierCode 2.0-AA against Ember’s 27.2, and the two tie at 36 on the Artificial Analysis Index. Ember leads on AutomationBench-AA, 37.8 against 13.9 for Haiku and 20.2 for Luna.
As we read it, the ladder is built for one workflow shape, a large model that plans and many small agents that repeat, with Meta pricing both halves.
Mark Zuckerberg, Meta CEO, on X, 9 November 2026
“our smallest model that delivers higher intelligence per dollar than others in its class”
Three Implications
FOR DEVELOPERS Meta’s developer blog says a team with an OpenAI-compatible client can point it at the Model API and set the model to muse-ember-1.0.
FOR ANTHROPIC AND OPENAI Their own pricing pages list Claude Haiku 5.5 and GPT-6 Luna at the same $0.10 input and $0.50 output as Ember, so Meta argues on cost per task, citing Artificial Analysis.
FOR META’S SALES PUSH The Information reported that Meta is forming a sales organization for small and midsize businesses; Meta has not said whether that team sells the Model API.
The framework argues: “an agent is a process that consumes tokens to produce work. The new job of finance is to manage the ratio between the two”.
As we read it, Meta’s pitch for Ember is that ratio: in its own example the cheapest model read 97% of the input tokens, by our arithmetic. The framework predates this launch and does not discuss Meta or Muse Ember.
What Is Not Established
Meta gives no usage, revenue or customer figures for the Model API. The partner quotes from Box, Cline, Oracle, Cloudflare and You.com describe early testing, not production deployments. Meta’s internal benchmarks, including its Personal Superintelligence Index, have not been reproduced independently, and Meta does not say whether the Muse Ember price is fixed or introductory.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Meta now sells three API tiers whose list prices span 40x, by our arithmetic, and has priced its smallest one level with Claude Haiku 5.5 and GPT-6 Luna. In Meta’s own example, Ember read 97% of the tokens for 46% of the bill, by our arithmetic, which is the work Meta says agents repeat all day.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The launch, the specifications, the benchmark table, the cost-per-task chart and the Python-to-Rust run come from Meta’s research and developer blogs of 9 November. Meta’s model prices come from its Muse model pages; rival prices come from Anthropic’s, OpenAI’s and Google’s own pricing pages, all read on 9 November. Benchmark results are Meta’s and Artificial Analysis’s as published by Meta. The price ratios and token shares are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
