Overview
Muse Ember is a model from Meta Superintelligence Labs, released on November 9, 2026. Meta calls it its smallest and most efficient model, designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction and long document parsing. It is the first model Meta trained using Muse Wonder Full, and it is meant to be paired with the larger Muse Wonder and Muse Spark models — the big model plans and orchestrates, Ember carries out the high-volume steps.
Ember natively understands text, images, audio and video, so an agent can read screenshots, charts and scanned documents, or work from voice memos, meeting recordings and video without a separate model. Meta says it can operate a computer directly from screenshots. It has a 1M-token context window and can produce up to 512K output tokens, enough to take in a full codebase or document set in one call.
Muse Ember (model id muse-ember-1.0) is served through the Meta Model API and Muse Code, and supports the Responses API. Below 200K tokens of context it costs $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens; above 200K the rates are $0.30, $0.012 and $1.50. Meta evaluated it under its Meta Superintelligence Scaling Framework and reports that it stayed below the framework's high-risk thresholds for chemical and biological, cybersecurity and loss-of-control risks.
| Released | 2026-11-09 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Parameters | Undisclosed |
| Context | 1M |
| Max output | 512K |
| Modalities | Text, Vision, Audio, Video |
| Status | Generally available |
Benchmarks
Muse Ember vs Claude Haiku 5.5 and GPT-6 Luna, each at max reasoning, as published by Meta
| Benchmark | Muse Ember | Claude Haiku 5.5 | GPT-6 Luna |
|---|---|---|---|
| AutomationBench-AA | 37.8% | 13.9% | 20.2% |
| Agents' Last Exam | 61.6% | 57% | 57.1% |
| Personal Superintelligence Index (Meta internal) | 46.8% | 36.1% | 37% |
| Terminal-Bench 4.0 | 40.9% | 39.2% | 16.4% |
| Coding Agent Index-AA | 46 | 37 | 31 |
| FrontierCode 2.0-AA | 27.2% | 34.9% | 25.2% |
| GDP.pdf-AA | 22.8% | 20.8% | 22.8% |
| OSWorld 2.1 (partial) | 69.5% | 68.9% | 50.5% |
| OSWorld 2.1 (binary) | 33.8% | 27.8% | 17% |
| PerceptionBench (with tools) | 80.6% | 78.6% | 62.8% |
| Artificial Analysis Intelligence Index | 36 | 36 | 28 |
This model's scores
- Agents' Last Exam61.6%
- AutomationBench-AA37.8%
- Terminal-Bench 4.040.9%
- OSWorld 2.1 (partial)69.5%
- PerceptionBench (with tools)80.6%
- Coding Agent Index-AA46
- Artificial Analysis Intelligence Index36
Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.
Pricing
| Input | $0.10 / 1M tokens |
|---|---|
| Cached input | $0.004 / 1M tokens |
| Output | $0.50 / 1M tokens |
muse-ember-1.0 on the Meta Model API below 200K tokens of context. Above 200K: $0.30 input / $0.012 cached / $1.50 output per 1M tokens.
Strengths
- Built for the many small, structured steps agents run: tool calls in sequence, coding subtasks, routing, extraction and document processing
- Native multimodal input — text, images, audio and video — including computer use from screenshots
- 1M-token context window with up to 512K output tokens
- Low token price: $0.10 input / $0.50 output per 1M tokens below 200K context, with cached input at $0.004
- Meta's published Artificial Analysis Intelligence Index of 36, against 37 for Muse Spark 1.3 at a fraction of the cost per task ($0.14 vs $2.39)
Best for
- Reach for it as the sub-agent in a harness: let Muse Wonder or Muse Spark plan, and send the repeated tool calls, file reads, test runs and targeted fixes to Ember.
- Reach for it for high-volume extraction and review over long documents, scanned pages, charts or recordings, where one multimodal model replaces a pipeline of separate ones.
- Reach for Muse Spark or Muse Wonder instead when a single hard problem matters more than cost per call.
How to access
| Provider | Model ID |
|---|---|
| Meta Model API ↗ | muse-ember-1.0 |
FAQ
When was Muse Ember released?
Meta released Muse Ember on November 9, 2026. It went live the same day on the Meta Model API and in Muse Code under the model id muse-ember-1.0.
How much does Muse Ember cost?
Below 200K tokens of context, Muse Ember costs $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens on the Meta Model API. Above 200K tokens the rates are $0.30 input, $0.012 cached and $1.50 output per million tokens.
What is Muse Ember used for?
Meta built Muse Ember for the work agents repeat all day — tool calls, coding subtasks, routing, extraction and long document parsing. It is meant to run alongside a larger model: Muse Wonder or Muse Spark plans and orchestrates, and Ember handles the volume of smaller steps.
What is Muse Ember's context window?
Muse Ember has a 1M-token context window and can produce up to 512K output tokens. It accepts text, image, audio and video input.
Is Muse Ember open weights?
No. Muse Ember is a proprietary, closed-weight model available through the Meta Model API and Muse Code. Meta's open-weight Muse models are Muse Glimmer and Muse Spark Open.