Meta has launched Muse Ember, the smallest and cheapest model in its Muse family of artificial intelligence models. The company announced the model on Monday 9 November 2026, and it is available now on the Meta Model API and inside Muse Code.

Ember is built for the work AI agents repeat all day, rather than for the hardest single questions. Meta says it is designed for tool calls, small coding tasks, routing requests, extracting data and reading long documents. The idea is that a larger model plans the job, while Ember handles the volume of smaller steps.

What Muse Ember costs and what it can do

Muse Ember costs $0.10 per million input tokens and $0.50 per million output tokens on the Meta Model API, where it is listed as muse-ember-1.0. It has a one-million-token context window and can produce up to 512,000 output tokens in a single run.

The model accepts text, images, audio and video as input. Meta says agents using it can work from screenshots, charts, scanned documents, voice notes, meeting recordings and video without switching to a different model. It also supports the Responses API.

Meta has not disclosed a parameter count for Ember, and its weights stay closed, unlike the open-weight Muse Glimmer model released in August. Early partners named by Meta include Box, Cline, Oracle, Cloudflare and You.com.

Meta's benchmark claims

Meta says Ember scores 36 on the Artificial Analysis Intelligence Index, the same score as Anthropic's Claude Haiku 5.5 at maximum reasoning, at about a third of the cost per task: $0.14 against $0.41. The company also claims Ember delivers more intelligence per dollar than Claude Haiku 5.5 or OpenAI's GPT-6 Luna at any reasoning setting.

Those figures are Meta's own claims, based on third-party evaluation data the company cites. No independent verification of the benchmark results has appeared yet, and the comparison depends on the test chosen. On at least one other test, the picture is mixed: Haiku outperforms Ember on real coding tasks in FrontierCode 2.0-AA, scoring 34.9 percent against Ember's 27 percent.

The fourth model in the Muse family

Ember is the fourth model released under the Muse brand from Meta Superintelligence Labs. It follows Muse Spark, released in April; the open-weights Muse Glimmer, released in August; and the flagship Muse Wonder, released in October. Meta says Ember is the first model trained using Muse Wonder Full, its largest model, and pitches the pair as a team: Wonder plans, Ember handles the volume.

The launch comes as Meta builds a dedicated sales team to sell AI products to business customers, working with consulting partners including Accenture and Deloitte. A low-cost model for high-volume agent work is central to that pitch, since businesses running agents thousands of times a day feel the price of every token.

Meta's consumer AI assistant, the Muse Agent, launched in September, and features such as Spaces and video chat have rolled out since. Ember sits behind the scenes rather than in front of users, but if Meta's cost claims hold up, it could become the workhorse model that powers many of those everyday agent tasks.