Meta has launched Muse Ember, its smallest and most efficient model, for the high-volume work that keeps AI agents running: tool calls, coding subtasks, routing, extraction, and long-document parsing. It is available now in Muse Code and through the Meta Model API as muse-ember-1.0, with Responses API support. Pricing is $0.10 per million input tokens, $0.004 per million cached input tokens, and $0.50 per million output tokens.
1/ announcing our smallest, most efficient little model is here: muse ember 🔥
— Alexandr Wang (@alexandr_wang) November 9, 2026
built for work that agents repeat all day, it gets you more intelligence per dollar than Haiku, Luna, and all other flash models at any reasoning setting and does it at a third of the cost or less! pic.twitter.com/ClmgPOTrsR
Ember has a 1 million-token context window and supports up to 512,000 output tokens, enough to take in a full codebase or document set in one call. Agents can use it to read files, run tests, make targeted fixes, pull records, call tools in sequence, and update systems across apps. It natively processes text, images, audio, and video, including screenshots, charts, scans, voice memos, meetings, and long documents.
Today we're introducing Muse Ember, our smallest and most efficient model, built for complex reasoning execution and everyday agent work.
— AI at Meta (@AIatMeta) November 9, 2026
Muse Ember can carry workflows across apps, call tools in sequence, run tests and make targeted code fixes. With a 1M-token context window… pic.twitter.com/qqcjZw6L8x
Meta claims Ember delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at every reasoning setting. At maximum reasoning, Ember reaches 35.8 on the Artificial Analysis Intelligence Index at an estimated $0.136 per task, compared with 36.2 at $0.411 for Haiku and 27.8 at $0.152 for Luna. The chart uses proposed Ember pricing and competitors' list prices. Ember also leads both rivals on AutomationBench-AA with 37.8 and Agents' Last Exam with 61.6, while its 27.2 on FrontierCode 2.0-AA trails Haiku's 34.9. Meta says no model on the efficiency frontier scores higher for less per task.
Meta's newest model Muse Ember scores 36 on the Artificial Analysis Intelligence Index, matching Claude Haiku 5.5 (max) at a third of the Cost per Task
— Artificial Analysis (@ArtificialAnlys) November 9, 2026
Muse Ember is Meta's newest efficient-tier model line, with the same input and output token prices as Claude Haiku 5.5 and… pic.twitter.com/ZXq4Zw8DZj
Ember is the first model trained using Muse Wonder Full. Meta positions Wonder for the hardest planning problems, Spark for coding and agent workflows, and Ember for repeatable execution at scale. Against Muse Spark 1.3 at maximum reasoning, Ember scores 36 versus 37 on the index while cutting estimated task cost from $2.39 to $0.14.
Early partners include Box, Cline, Oracle, Cloudflare, and You.com. Box tested Ember on contract risk scoring and candidate screening, while Cline sees it as a low-cost option for coding subagents and document analysis. Meta says pre-release reviews placed Ember below high-risk thresholds in assessed chemical and biological, cybersecurity, and loss-of-control categories. Safety testing also covered sensitive data, prompt injection, user intent, and seeking human input before high-impact actions.
Sources and related context
- Meta introduces Muse Wonder and Muse Wonder Full: Related context: Meta's earlier release explains the Wonder family that it says was used to train Muse Ember.
- Meta introduces Muse Spark 1.3: Related context: This release describes the coding and agent workflows supported by Spark 1.3, the model Meta compares with Ember.
