Google Cloud's documentation now lists Muse Ember 1.0 from Meta as a managed API on its Gemini Enterprise Agent Platform, and TechCrunch reported that Meta Superintelligence Lab released the model on November 9, 2026. Google Cloud's Ember model card lists the launch stage as Preview. As of November 10, 2026, Google Cloud's pricing page lists Ember at $0.10 per 1M input tokens and $0.50 per 1M output tokens for prompts up to 200K tokens.

#What is Muse Ember 1.0?

Ember is Meta's small model for repeated agent work, and Google pitches it for volume rather than planning. Google Cloud's Meta models page says: "Muse Ember 1.0 from Meta is an efficient reasoning model built for high-volume agent tasks, including tool calls, coding subtasks, routing, extraction, and long-document processing."

TechCrunch calls it "the lab’s lightest and most cost-efficient AI model yet." According to TechCrunch, Meta said Ember was "designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing". TechCrunch says Ember is the fourth model released under Meta's Muse brand.

#What are the specs?

Google Cloud's model card sets Ember's limits and inputs:

Spec

Muse Ember 1.0 on Google Cloud

Model ID

meta/muse-ember-1.0

Release date

November 9, 2026

Launch stage

Preview

Maximum input tokens

1,048,576

Maximum output tokens

524,288

Inputs

Text, image

Outputs

Text

Supported

Function calling, structured output

Not supported

Batch predictions, Provisioned Throughput

Availability

Global endpoint

Preview offerings fall under Google's Pre-GA Offerings Terms, and the card says they "might have limited support."

Meta claims broader media skills than Google lists. TechCrunch reports that "Meta also trained Ember to work well across different types of media, including text, images, and audio", and that agents can work from voice memos, meeting recordings and video without a separate model. Google Cloud's card lists only text and image inputs with text output, so audio and video are Meta's claim, not a documented Google Cloud feature.

#How much does it cost?

Google Cloud's partner pricing page lists these Ember rates in US dollars per 1M tokens, as of November 10, 2026:

Usage

Up to 200K input tokens

Over 200K input tokens

Input

$0.10

$0.30

Output

$0.50

$1.50

Cache hit

$0.004

$0.012

Google says queries over 200K input tokens "are charged at long-context rates." The same page lists Muse Spark 1.3 at $1.25 input and $4.25 output, and Muse Wonder 1.0 at $4.00 input and $20.00 output, per 1M tokens. Google Cloud's card lists standard pay-as-you-go usage for Ember.

TechCrunch reports that Meta used Artificial Analysis's composite benchmark score to show Ember as cheaper and smarter than Anthropic's Haiku and OpenAI's Luna. That is Meta's claim; our sources do not publish the scores.

#Who can use it and how?

Ember is available through Google Cloud's Agent Platform. Google's docs give the setup:

  1. Go to the Meta model's card in Model Garden and enable it.
  2. Call the stateless, OpenAI-compatible Responses API on the global endpoint.
  3. Pass the bare model ID muse-ember-1.0 in the model field.
  4. Use server-sent events (SSE) if you want streamed responses.

Stateful response storage (store: true) is not supported, Google says. Its terms also note that Meta models are not a Google product.

#How does it compare?

Ember is built to sit under a bigger planner. Google's docs say: "Pair it with Muse Wonder 1.0 or Muse Spark 1.3 for multi-step planning and reasoning." TechCrunch quotes Meta's line: "Muse Wonder plans, Muse Ember handles the volume." We covered Wonder in Meta launches Muse Wonder and Muse Wonder Full models for agents, and Meta's consumer agent updates in Meta Muse adds custom spaces, voice and expressive avatars.

Our analysis: On Google Cloud's listed rates, Ember's input price for prompts up to 200K tokens is one fortieth of Wonder's, and its output price one fortieth too. That gap fits the pairing Meta describes, where Wonder plans and Ember handles the volume. Teams should weigh the Preview label first: Pre-GA terms and possible limited support make it a fit for testing, not yet for critical production work.

#What is still unclear?

No Meta announcement page was available in our sources, so Meta's own specs and benchmark figures are not cited here. Google Cloud has not said when Ember will leave Preview. Audio and video support rests on Meta's claim via TechCrunch, and Google's card does not list them.