Credit: Meta Meta has released Muse Ember, its smallest and cheapest AI model so far, the company announced on Monday. Meta built it for the small jobs AI agents repeat all day. These include calling tools, coding subtasks, routing requests, pulling out data and parsing long documents. Ember costs $0.10 per million input tokens and $0.50 per million output tokens.
Its two closest rivals, Anthropic’s Claude Haiku 5.5 and OpenAI’s GPT-6 Luna, have the same list prices. Meta says Ember gets more done for each dollar than either of them, at every reasoning setting. Chief executive Mark Zuckerberg announced the launch on X. Releasing Muse Ember today — our smallest model that delivers higher intelligence per dollar than others in its class.
Built for token-heavy work that agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing. Available today on the Meta Model API and in Muse Code. – Mark Zuckerberg (@finkd) November 9, 2026 A third of the cost per task The cost claim rests on the Intelligence Index from Artificial Analysis, an independent benchmarking firm. Ember scores 36 on its highest setting, the same as Haiku 5.5 on its highest setting. Luna scores 28.
Running the index cost about $0.14 per task with Ember, according to Meta’s figures. It cost $0.41 for Haiku and $0.15 for Luna. That puts Ember level with Haiku at about a third of the cost. “Our teeny model muse ember is now on the efficient frontier, matching haiku at 3x cheaper cost per task, and smarter than luna,” Meta’s AI chief Alexandr Wang wrote on X. Ember also leads both rivals on Meta’s other agent and coding tests.
On AutomationBench, which tests full business workflows, it scored 37.8%, against 20.2% for Luna and 13.9% for Haiku. It does not win everywhere. On FrontierCode 2.0, which checks whether maintainers would merge a model’s code changes, Haiku scored 34.9% and Ember 27.2%. Several of the other results, including Meta’s Personal Superintelligence Index, come from Meta’s own internal tests.
Built to work under a bigger model Ember can take in text, images, audio and video. It can read up to a million tokens at once, enough for a full codebase. Meta pitches it as a partner for Muse Wonder, its frontier model. “For the hardest problems, pair it with Muse Wonder: Muse Wonder plans, Muse Ember handles the volume,” Meta wrote. Ember is also the first model Meta trained using the larger Muse Wonder Full.
Meta is keeping Ember’s model weights private, as with its previous Muse releases, Frederic Lardinois reported for The New Stack. In October, Meta released a weaker open model while keeping its frontier models closed. Ember’s input price also undercuts Chinese rivals, according to The New Stack. They include DeepSeek V4.1 Flash, Z.ai’s GLM-5.3 Flash and MiniMax M3.1 Flash.
All three score lower on the Artificial Analysis index. Meta tested Ember against its own safety framework, covering chemical and biological weapons, cybersecurity and loss of control. It stayed below the framework’s high-risk thresholds in every category, the company said. Ember is available now through the Meta Model API and Muse Code, Meta’s coding tool.
It joins a Muse line-up that already powers Meta’s Muse agent. Meta said it has even more efficient models in training. Get the TNW newsletter ✕ Close Get the TNW newsletter Get the most important tech news in your inbox each week.
Leave a Reply