Meta launches Muse Ember, a low-cost model for AI agents

Meta Superintelligence Labs released Muse Ember on Monday, Nov. 9. The company calls it its lightest and most cost-efficient AI model so far. According to TechCrunch, it is "designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing." The launch gives Meta a direct rival to Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6 Luna in the market for cheap, high-volume models that do the repetitive work inside AI agent systems.techcrunch.com

Pricing and benchmarks

Ember costs $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens. It has a 1-million-token context window and can produce up to 512,000 output tokens, The New Stack reported. Those input and output prices match what Luna and Haiku charge for standard short prompts. Meta's pitch rests instead on the cost of finishing a task. At maximum reasoning, Meta puts Ember at about $0.14 per task on the Artificial Analysis Intelligence Index, compared with $0.41 for Haiku and $0.15 for Luna. On that index Ember scores 36, the same as Haiku and above Luna's 28. TechCrunch noted that the model's prices are also competitive with open-source rivals such as DeepSeek V4.1 Flash and GLM 5.3 Flash.thenewstack.io+1

The results were not all in Ember's favor. On FrontierCode 2.0, which tests whether a repository maintainer would merge an agent's code changes, Ember scored 27.2%, behind Haiku at 34.9%. On the OSWorld 2.1 computer-use benchmark, it completed 33.8% of tasks outright, compared with 27.8% for Haiku and 17% for Luna.thenewstack.io

Built to pair with Muse Wonder

Ember is meant to work inside agent systems alongside Meta's larger Muse Wonder. In Meta's announcement, as quoted by TechCrunch, the company sums up the split this way: "Muse Wonder plans, Muse Ember handles the volume." Ember accepts text, images, audio and video. Meta says agents "can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model." It is also the first model Meta trained using its larger Muse Wonder Full model.techcrunch.com+1

Developers can use the model in Muse Code and through the Meta Model API under the ID muse-ember-1.0. As with Meta's earlier Muse releases, the model weights are not public. Meta's developer blog describes Ember as built for "high-volume execution and multimodal inputs."thenewstack.io+1

A fast-growing model family

Ember is the fourth model under the Muse brand. Before it came the mid-weight Spark in April, its open-source counterpart Glimmer in August and the frontier model Wonder in October. Meta's consumer Muse Agent launched in September, and new features such as Spaces and video chat were added earlier this month.techcrunch.com