New to this? Read it in simple words
- Meta released Muse Ember, its smallest model, on November 9.
- An outside test scored it 36, the same as Claude Haiku 5.5 (max).
- A test task cost $0.14 with Ember, against $0.41 with Haiku 5.5.
- It declines 64% of questions on one knowledge test, so it knows less.
- Cost per task
- What it costs a model to finish one whole job, not just one piece of text.
- Cached input
- Text a model has already read once, which services charge less for when it is read again.
- Context window
- How much text a model can take in at once.
Muse Ember is the newest model in Meta’s Muse family, after the larger Muse Wonder in October, TechCrunch reports. Meta builds the two to work together on agent jobs.
What it is
Meta calls Muse Ember its “smallest and most efficient model”. It is built for tool calls, coding subtasks, routing, extraction and reading long documents.
It reads text, images, audio and video, and writes text. Its context window, the amount it can take in at once, is 1 million tokens. It runs on the Meta Model API and in Muse Code, Meta’s coding tool.
Artificial Analysis lists it as a closed model, so its weights are not public. Meta plans to pair it with its bigger model. “Muse Wonder plans, Muse Ember handles the volume,” Meta says, as TechCrunch reports.
All figures are from Artificial Analysis’s launch-day test.
The outside test
Artificial Analysis scored Muse Ember (max) at 36 on its Intelligence Index, the same as Claude Haiku 5.5 (max). Ember cost $0.14 per test task, and Haiku 5.5 cost $0.41. At xhigh effort, Ember scored 34 for $0.12.
Its prices per token are the same as those of Haiku 5.5 and GPT-6 Luna, the test group says. But Ember’s cached input, text it has already read once, is 60% cheaper. It also writes about a third as many output tokens as Haiku 5.5.
Meta’s bigger Muse Wonder (max) scores 44, but it costs $6.95 per task.
Coding and facts
Ember trailed Haiku 5.5 on FrontierCode 2.0, a coding test, with 27.2% against 34.9%. In Meta’s own coding tool, it scored 46 on the test group’s coding agent index. Haiku 5.5 in Claude Code scored 37, and GPT-6 Luna in Codex 31.
Knowledge is its main weakness. On AA-Omniscience, a test of facts, Ember answered 21.1% of questions correctly, against 34.7% for Haiku 5.5 (max).
That is mostly because it declines to answer 64% of the questions, the test group says. In return, it rarely makes things up: its rate of made-up answers is 17.4%, half of Haiku 5.5’s.
Sources6
Meta’s own claims
Meta says Ember gives more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna “at any reasoning setting”. The outside test covers the Haiku comparison; the claim about every Luna setting is Meta’s.
Meta also says Ember stayed below the high-risk thresholds of its safety framework. That is Meta’s own finding.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2
- 3Research · November 9, 2026Meta releases Muse Ember, a new lightweight model for agent execution TechCrunch
- 4Research · November 9, 2026Meta launches Muse Ember, a small AI model built for the repetitive work agents actually do The Decoder
- 5Research · November 9, 2026Meta’s Muse Ember matches Claude Haiku at a third of the cost per task The Next Web
- 6Research · November 9, 2026Meta’s newest model Muse Ember scores 36 on the Artificial Analysis Intelligence Index, matching Claude Haiku 5.5 (max) at a third of the Cost per Task Artificial Analysis
The scores and task costs come from Artificial Analysis, an outside test group. The claim about every GPT-6 Luna setting and the safety result are Meta’s own.
Meta’s Muse Ember matches Haiku 5.5 at a third of the cost. Silicon AI News, November 9, 2026. https://siliconainews.com/stories/meta-muse-ember-small-model
Meta’s smallest model matches Claude Haiku 5.5’s outside score at a third of the task cost, though it knows less. How we rate the mood
The mood shows how the news frames the event for the tech and AI world: as an opportunity (6 and above) or as a loss (below 5). It is not our verdict on who is right.