Meta launched Muse Ember on Monday, November 9, calling it the smallest and most efficient model in its Muse lineup. The new model costs $0.10 per million input tokens and $0.50 per million output tokens, and it handles a context window of 1 million tokens. Meta built it for one job: the repetitive work AI agents do all day.

That work is specific. Muse Ember handles tool calls, coding subtasks, routing, extraction, and parsing long documents, the chores that pile up when an agent runs for hours. Mark Zuckerberg announced it on X, writing, “Releasing Muse Ember today — our smallest model that delivers higher intelligence per dollar than others in its class.”

The pitch is simple: big models plan, Ember does the volume.

What Meta Muse Ember can do

Ember reads text, images, audio, video, and PDFs, then answers in text. Meta says agents can work from screenshots, charts, scanned documents, voice memos, and meeting recordings without switching to a separate model. It can return up to 512,000 tokens in a single response, which makes it suited to jobs that chew through long files.

Developers can call it today through the Meta Model API and Muse Code under the model ID muse-ember-1.0. Meta also lists it on OpenRouter, Google Cloud, and Oracle Cloud. Early partners named by Meta include Box, Cline, Oracle, Cloudflare, and You.com.

The price math behind Meta Muse Ember

Ten cents per million input tokens is the headline number, but the fine print matters. Cached input costs $0.004 per million tokens, a 96% discount on repeated prompts, and both rates triple once a request passes 200,000 input tokens. For agents that reuse the same long instructions all day, that cache price is the number that counts.

Meta frames the value as cost per finished task, not cost per token. On the Artificial Analysis Intelligence Index, Ember scored 36 at its maximum reasoning setting, matching Claude Haiku 5.5, while costing about $0.14 per task against Haiku’s $0.41. Muse Spark 1.3, Meta’s bigger coding model, scored 37 but cost $2.39 per task on the same test.

Meta Muse Ember vs Claude Haiku 5.5

Meta’s comparison holds up on the composite index, but it gets thinner on real coding. The Decoder reported that Haiku 5.5 still wins on FrontierCode 2.0-AA, a real-world coding test, scoring 34.9% to Ember’s 27.2%. Meta’s own docs steer hard coding and knowledge-heavy questions to its larger models, Wonder and Spark, and tell developers not to use Ember for questions answered from memory.

GPT-6 Luna, OpenAI’s small model, is the other name in Meta’s comparison. Meta says Ember delivers more intelligence per dollar than Haiku or Luna at every reasoning setting, though those are Meta’s own numbers and the company picked the tests.

Muse also lands on Apple Watch

Ember wasn’t Meta’s only launch on Monday. Its Muse AI assistant arrived on Apple Watch and Wear OS devices the same day, letting users type, dictate, or call the assistant from their wrist. Apple Watch support needs watchOS 9.0 or later, and a Ray-Ban complication puts Muse on the watch face. For now the watch app is limited to the US, Canada, Singapore, and Japan.

Why this matters

Ember is the fourth Muse model Meta has shipped this year, after Spark in April, the open-source Glimmer in August, and the flagship Wonder in October. The lineup now reads like a division of labor: Wonder plans, Spark codes, Ember grinds through volume. Meta even trained Ember using Muse Wonder Full, its biggest model, so the flagship built its own smaller sibling.

That structure is the story. AI agents burn most of their tokens on small repeated steps, not on the hard reasoning, and a cheap model that handles those steps changes what an agent run costs. If Meta’s per-task math holds up outside its own tests, agent apps get cheaper to run. If it doesn’t, Ember is still a useful reminder that the model race is now fought on price per job, not just on benchmark scores.

Frequently asked questions

What is Meta Muse Ember?

Meta Muse Ember is Meta’s smallest AI model, launched November 9, 2026. It is built for the repetitive work AI agents do all day, including tool calls, coding subtasks, routing, extraction, and parsing long documents.

How much does Meta Muse Ember cost?

Muse Ember costs $0.10 per million input tokens and $0.50 per million output tokens. Cached input costs $0.004 per million tokens, but both rates triple once a request passes 200,000 input tokens.

How does Muse Ember compare with Claude Haiku 5.5?

Both scored 36 on the Artificial Analysis Intelligence Index, but Ember cost about $0.14 per task against Haiku’s $0.41. Haiku still wins on real coding work, scoring 34.9% to Ember’s 27.2% on FrontierCode 2.0-AA.

Where can developers use Meta Muse Ember?

Muse Ember is available through the Meta Model API and Muse Code under the model ID muse-ember-1.0. It is also listed on OpenRouter, Google Cloud, and Oracle Cloud.

Sources: TechCrunch, The Decoder, Economic Times