Meta 发布小模型 Muse Ember，面向智能体日常重复任务
Meta launches Muse Ember, a small AI model built for the repetitive work agents actually do
Meta 发布其最小、最高效的模型 Muse Ember，面向智能体日常任务，包括调用工具、小型编码、请求路由、数据抽取和长文档处理。该模型支持文本、图像、音频和视频，单次可处理最多一百万 token，已通过 Meta Model API 和 Muse Code 开放，定价为每百万输入 token 0.10 美元、每百万输出 token 0.50 美元。Meta 称其在各推理档位上每美元智能都高于 Claude Haiku 5.5 和 GPT-6 Luna，在 Artificial Analysis 测试中每任务成本约 0.14 美元，而更大的 Muse Spark 1.3 为 2.39 美元、得分 36 对 37；但在 FrontierCode 2.0-AA 的真实编码任务上 Haiku 以 34.9% 领先 Muse Ember 的 27.2%。
推荐理由
AI 评分 68
原文给出每任务成本与 FrontierCode 2.0-AA 上的对比结果，可据此判断小模型在智能体日常任务中的性价比边界。
主题
正文
Meta has introduced Muse Ember, its smallest and most efficient AI model to date. It's designed for the routine tasks AI agents handle all day: calling tools, handling small coding tasks, routing requests, extracting data, and processing long documents. Meta says Muse Ember delivers more intelligence per dollar at every reasoning level than Claude Haiku 5.5 or GPT-6 Luna. It costs about $0.14 per task in the Artificial Analysis test , while the larger Muse Spark 1.3 costs $2.39 for a nearly identical score of 36 versus 37. Muse Ember leads in most of Meta's own tests, but Haiku outperforms it on real coding tasks in FrontierCode 2.0-AA, scoring 34.9 percent to Muse Ember's 27.2 percent.
The model handles text, images, audio, and video and can process up to one million tokens at once. It's available now through the Meta Model API and Muse Code at $0.10 per million input tokens and $0.50 per million output tokens.