Meta Muse Ember Targets AI Agents at About $0.14 per Task

Meta Superintelligence Labs launched Muse Ember on Monday, Nov. 9, 2026, calling it the company's smallest and most efficient AI model. According to Meta's announcement, Muse Ember is built for "the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing," and it costs about $0.14 per task on the Artificial Analysis benchmark.

The model is available today in Muse Code and the Meta Model API under the model ID muse-ember-1.0, with support for the Responses API. It natively reads text, images, audio and video, so an agent can work from screenshots, charts, scanned documents, voice memos or meeting recordings without a separate model. Meta is keeping the weights private, as it did with earlier Muse releases.

How much does Muse Ember cost?

Meta lists Muse Ember at $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens. Tokens are the chunks of text a model reads and writes, and they are how AI APIs bill usage. The model has a 1 million token context window and can return up to 512,000 tokens in one response.

Those headline rates are not a price cut. The New Stack points out that they match GPT-6 Luna and Claude Haiku 5.5 at their standard short context prices. Meta's real claim is a lower cost per finished task, which depends on how many tokens a model burns to get the job done.

How does Muse Ember compare with Claude Haiku 5.5 and GPT-6 Luna?

Meta says Ember delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at every reasoning setting. At max reasoning, Ember scores 36 on the Artificial Analysis Intelligence Index at about $0.14 per task, while Haiku 5.5 also scores 36 but costs about $0.41 per task. GPT-6 Luna scores 28 at about $0.15. Anthropic's bigger models are a different tier; we covered them in our look at Claude Fable 5.2 and Mythos 5.2.

The benchmark table below uses Meta's published figures at max reasoning. Some of them come from Meta's internal evaluations, and Haiku still wins the FrontierCode test.

Benchmark	Muse Ember	Claude Haiku 5.5	GPT-6 Luna
AutomationBench-AA	37.8	13.9	20.2
Terminal-Bench 4.0	40.9	39.2	16.4
Coding Agent Index-AA	46	37	31
FrontierCode 2.0-AA	27.2	34.9	25.2
OSWorld 2.1 (partial credit)	69.5	68.9	50.5
Cost per Intelligence Index task	about $0.14	about $0.41	about $0.15

Why is Meta building a small model for agents?

Meta pitches Ember as a subagent, a smaller model that handles the tasks a bigger planner model hands off. In Meta's words, "Muse Wonder plans, Muse Ember handles the volume." Ember is also the first model Meta trained using Muse Wonder Full, and it follows the Muse models Meta began rolling out to Meta AI users earlier this month.

Early testers include Box, Cline, Oracle, Cloudflare and You.com. Cloudflare CTO Dane Knecht said Ember "ran up to 30% faster at the tail than the comparable model we tested," and You.com reported a p95 latency of 2.2 seconds in its coding tests.

On safety, Meta says Ember stayed below the high risk thresholds of its Superintelligence Scaling Framework for chemical, biological, cybersecurity and loss of control risks. It was also tested against prompt injection, where hidden instructions in a web page or tool output try to hijack an agent.

What does Muse Ember mean for developers?

For teams running AI agents at scale, the cached input price of $0.004 per million tokens and the 1 million token window matter more than the list price. Repetitive jobs that reuse the same long instructions or documents could get much cheaper.

Our take: Ember is a credible attempt to win the subagent slot that Haiku and Luna hold today, but the cost per task advantage rests on Meta's own and Artificial Analysis numbers. Developers should run it on their own workloads before switching, especially for hard coding work where Haiku still leads.