正文 · AI 翻译
IT Home reported on November 10 that Meta has launched the Muse Ember AI model. It is Meta's smallest and most efficient model to date, designed for complex reasoning tasks and the work that agents handle repeatedly every day.
Meta said the model performs comparably to Claude Haiku 5.5 on the Artificial Analysis leaderboard, but costs just 1/3 as much per task and surpasses GPT-6 Luna on several tests.
IT Home provides the Muse Ember benchmark results below:
Benchmark
Muse Ember (max)
Claude Haiku 5.5 (max)
GPT-6 Luna (max)
Agents: AutomationBench-AA (end-to-end business workflows)
37.8
13.9
20.2
Agents' Last Exam (end-to-end agent tasks)
61.6
57.0
57.1
Personal Superintelligence Index (internal composite index for personal and knowledge work)
46.8
36.1
37.0
Coding: Terminal-Bench 4.0 (agent-based terminal coding)
40.9
39.2
16.4
Coding Agent Index-AA (coding in real-world agent environments)
46
37
31
FrontierCode 2.0-AA (real-world engineering tasks assessed by whether the code can be merged)
27.2
34.9
25.2
Multimodal: GDP.pdf-AA (professional work with long documents)
22.8
20.8
22.8
OSWorld 2.1 (computer use across desktop applications)
69.5
68.9
50.5
PerceptionBench (fine-grained visual perception)
80.6
78.6
62.8
Overall: Artificial Analysis Index
36
36
28
The model also has a 1 million-token context window and supports output of up to 512,000 tokens, allowing it to process an entire codebase or a full set of documents in a single call. It enables agents to perform hundreds of coding subtasks in one session, including reading files, running tests, and making targeted fixes.
Muse Ember natively understands text, images, audio, and video. Without relying on other models, agents can read screenshots, charts, and scanned documents, as well as process voice memos, meeting recordings, and videos. It can handle lengthy real-world documents, operate a computer directly from screenshots, and identify subtle visual details.
On safety, the model was evaluated under Meta's SI Scaling Framework before release. All test results showed that Muse Ember's risk level was below the framework's high-risk threshold.
On pricing, the model is available starting today through the Meta Model API, Google Cloud, and Oracle Cloud. Input costs $0.10 per million tokens (IT Home notes that this is about 0.67 yuan at current exchange rates), while output costs $0.50 per million tokens (about 3.3 yuan at current exchange rates).