Meta 发布轻量模型 Muse Ember，面向智能体执行场景
Meta releases Muse Ember, a new lightweight model for agent execution
- 1Meta 超级智能实验室发布轻量模型 Muse Ember，定位智能体全天的执行类任务。
- 2定价每百万输入 token 10 美分，官方称比 Haiku 和 Luna 更便宜且更智能。
- 3官方建议与 Muse Wonder 搭配，大模型定规则、Ember 批量执行数万案例。
- 4模型支持文本、图像、音频输入，官方称可直接读取截图、扫描文档和会议录音。
Meta 把竞争焦点放在智能体执行环节的单位任务成本上，用轻量模型承接大模型定好的批量任务。
- 影响：智能体执行类任务的单次成本有望下降，客服、文档处理等高频场景预算压力减轻；大小模型分工架构成为可参考的部署方式；小模型价格竞争也给模型采购议价留出空间。
- 参考：可考虑把客服、文档解析等批量环节按单任务成本重新估价，对比 Ember 与现有方案的报价；值得评估大模型定规则、小模型跑量的分工是否适配现有智能体流程。
- 依据：价格与产品定位为官方口径；更便宜更智能的对比引用 Artificial Analysis 综合基准，属 Meta 宣传，暂无独立验证。
Meta Superintelligence Lab released a new AI model on Monday called Muse Ember, the lab’s lightest and most cost-efficient AI model yet. In its release, Meta said Ember was “designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing.”
Meta particularly emphasized Ember’s cost-per-task, using Artificial Analysis’s composite benchmark score to show Ember as cheaper and smarter than competitors like Anthropic’s Haiku and OpenAI’s Luna.
At 10 cents per million input tokens, Ember is significantly cheaper than frontier competitors like Haiku and Luna. Not only that, it’s cost-competitive with open-source competitors like DeepSeek V4.1 Flash and GLM 5.3 Flash.
Given the emphasis on low cost and accuracy at scale, Ember is designed to pair with a larger model like Meta’s Muse Wonder in an agentic harness. In the case of a customer-support agent, for instance, you might see a heavier model establishing rules for how to handle a particular issue, then sending a lighter-weight model to apply the rules across tens of thousands of cases. As Meta puts it in the announcement, “Muse Wonder plans, Muse Ember handles the volume.”
With volume-heavy use cases in mind, Meta also trained Ember to work well across different types of media, including text, images, and audio. The result makes the model surprisingly flexible in what kind of inputs it can handle. As Meta puts it in the announcement, “agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model.”
Ember is the fourth model released under Meta’s Muse brand, following the mid-weight Spark (released in April), Spark’s open-source twin Glimmer (released in August), and the Wonder frontier model (released in October). The flagship consumer AI product, the Muse Agent, was released in September to unexpected success, with new features like Spaces and video chat rolling out earlier this month.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Russell Brandom has been covering the tech industry since 2012, with a focus on platform policy and emerging technologies. He previously worked at The Verge and Rest of World, and has written for Wired, The Awl and MIT’s Technology Review. He can be reached at [email protected] or on Signal at 412-401-5489.
来源：TechCrunch · AI · techcrunch.com