Muse Ember, released by Meta on November 9, 2026, is the smallest and cheapest model in the Muse family. Sold as muse-ember-1.0, it costs $0.10 per million input tokens, reads text, images, video, audio and PDFs, and has a 1M-token context window. It is built for sub-agents, tool calls and routing.
Last updated: November 9, 2026
Table of contents
- What Muse Ember is and what is new
- Specs, pricing and limits
- Muse Ember vs. Claude Haiku 5.5 and GPT-6 Luna
- How to make your first Muse Ember call
- Workflows: pairing Ember with a larger model
- Limitations
- Frequently asked questions
- Sources
What Muse Ember is and what is new
Meta now sells three hosted reasoning models: Muse Wonder for the hardest problems, Muse Spark for coding and agentic workflows, and Muse Ember for repeatable work at scale. The Muse Ember announcement describes it as a model for “the work agents repeat all day”: tool calls, coding subtasks, routing, extraction and long document parsing. Meta says it is the first model trained using Muse Wonder Full.
The key points for power users:
- Lowest price in the Muse lineup. $0.10 input and $0.50 output per million tokens, compared with $1.25 and $4.25 for Muse Spark 1.3 in Standard mode.
- Same request format as Muse Spark. The Muse Ember documentation says switching most examples is a change to the model ID.
- Full multimodal input. Text, image, video, audio and PDF in; text out.
- Agent features included. It works with computer use, image understanding, tool calling, structured output and search grounding.
- Three API shapes. Responses, Chat Completions and Messages endpoints are all supported.
If you followed Meta’s earlier releases, our explainer on Meta Muse Spark 1.3 and our guide to the open-weight Muse Spark Open cover the larger models Ember is meant to work alongside.
Specs, pricing and limits
All figures below come from Meta’s developer documentation, including the pricing and rate limits page.
| Item | Muse Ember (muse-ember-1.0) |
|---|---|
| Context window | 1,048,576 tokens |
| Maximum output | 524,288 tokens |
| Input types | Text, image, video, audio, PDF |
| Reasoning effort | “minimal” to “max”; default “max” |
| Price, up to 200K input tokens | $0.10 input, $0.004 cached input, $0.50 output per 1M tokens |
| Price, over 200K input tokens | $0.30 input, $0.012 cached input, $1.50 output per 1M tokens |
| Rate limits (per team) | 3,000 requests and 4,000,000 tokens per minute |
| Pricing modes | Standard mode only; no Contributor-mode variant |
Two billing details matter. First, the long-context tier applies to the entire request, not only the tokens past 200K, and cached tokens count toward the threshold. A request that continues a stored conversation with previous_response_id includes that history in its input. Second, the web_search grounding tool adds $2.50 per 1,000 search queries on top of token costs.
In Muse Code, Ember is included on every subscription plan, which start at $5 per month. Meta notes that requests over 200K input tokens use a plan’s allowance faster.
Muse Ember vs. Claude Haiku 5.5 and GPT-6 Luna
Meta compares Ember with the small models from Anthropic and OpenAI. All three run at max reasoning. Some scores come from Artificial Analysis and others from Meta’s internal evaluations, so treat them as vendor-reported.
| Benchmark | Muse Ember | Claude Haiku 5.5 | GPT-6 Luna |
|---|---|---|---|
| AutomationBench-AA (business workflows) | 37.8 | 13.9 | 20.2 |
| Terminal-Bench 4.0 (terminal coding) | 40.9 | 39.2 | 16.4 |
| Coding Agent Index-AA | 46 | 37 | 31 |
| FrontierCode 2.0-AA (mergeable code) | 27.2 | 34.9 | 25.2 |
| OSWorld 2.1, partial credit (computer use) | 69.5 | 68.9 | 50.5 |
| Artificial Analysis Intelligence Index | 36 | 36 | 28 |
Meta also puts Ember next to its own Muse Spark 1.3: an Intelligence Index score of 36 against 37, at about $0.14 per index task compared with $2.39. In other words, Meta is positioning Ember as nearly Spark-level on that index at a small fraction of the cost.
How to make your first Muse Ember call
Model API is compatible with the OpenAI and Anthropic SDKs, so most stacks only need a base URL and key change.
- Create a Model API account, then open the dashboard and choose API keys, then Create API key.
- Store the key as an environment variable, for example
export MODEL_API_KEY="your-key"on macOS or Linux. - Point your client at
https://api.meta.ai/v1. With the OpenAI SDK, pass the key explicitly, because the SDK does not readMODEL_API_KEYon its own. - Send a request to the Responses endpoint with
"model": "muse-ember-1.0". Meta’s own example classifies a support ticket as billing, bug or feature request. - Lower the reasoning effort from the default “max” for simple routing or extraction steps where latency and output tokens matter more than depth.
- For Anthropic-format tools such as Claude Code, use the Messages API pointed at
https://api.meta.aiand map the model aliases as described below.
Workflows: pairing Ember with a larger model
Meta’s model selection guide recommends a split: a larger model plans and Ember handles the volume. Practical setups it documents include:
- Claude Code and the Claude Agent SDK. Map the opus alias to
muse-wonder-1.0, the sonnet alias tomuse-spark-1.3, and the haiku alias and sub-agent model tomuse-ember-1.0. Our Claude Code 2.1.318 guide covers the client side. - Your own agent loop. Pass a different
modelvalue per request. The three reasoning models accept the same request body. - Bulk jobs. Run the same task over thousands of images, files or records, such as captioning photos, extracting fields or classifying tickets, with many Ember requests in parallel.
A good prompt pattern for Ember sub-agents is a narrow, bounded instruction with a fixed output schema. Pair it with structured output so routing and extraction results are guaranteed to be valid JSON. If you route across vendors instead, compare our guide to OpenRouter Auto Code.
Limitations
- Not for the hardest tasks. Meta advises using Muse Spark or Muse Wonder for hard single-shot coding problems, knowledge-heavy questions and visual coding.
- Long-context cost jump. Muse Wonder and Muse Spark charge one rate at any length, while Ember triples its rates above 200K input tokens. Compare costs at your real input sizes. For context limits elsewhere, see the Gemini context window.
- No Contributor mode. Unlike Muse Spark, there is no discounted variant that shares data to improve Meta’s products.
- Regional availability. Meta says model availability can vary by country under its Geographic Use Policy.
For Meta’s consumer-side updates, see our guide to Muse Spaces.
Frequently asked questions
What is Muse Ember?
Muse Ember is Meta’s smallest and most efficient Muse model, launched November 9, 2026. It is a reasoning model for high-volume agent work such as sub-agent tasks, tool calls, routing and extraction. It is available on Meta Model API and in Muse Code as muse-ember-1.0.
How much does Muse Ember cost?
For requests with up to 200K input tokens, Meta charges $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens. Requests over 200K input tokens are billed in full at $0.30, $0.012 and $1.50.
What is the Muse Ember context window?
Muse Ember has a context window of 1,048,576 tokens and a maximum output of 524,288 tokens, according to Meta’s developer documentation. Long prompts above 200K input tokens move the whole request to the higher long-context price tier.
Can I use Muse Ember in Claude Code?
Yes. Meta’s documentation says Claude Code connects to Model API through the Messages API, and recommends mapping the haiku alias and the sub-agent model to muse-ember-1.0, with larger Muse models on the opus and sonnet aliases.
Is Muse Ember better than Claude Haiku 5.5?
On Meta’s published results at max reasoning, Muse Ember leads Claude Haiku 5.5 on AutomationBench-AA, Terminal-Bench 4.0 and Coding Agent Index-AA, ties on the Artificial Analysis Index at 36, and trails on FrontierCode 2.0-AA. These are vendor-reported figures.
Sources
- Meta: Introducing Our Smallest, Most Efficient Model: Muse Ember (November 9, 2026)
- Meta Model API documentation: Muse Ember
- Meta Model API documentation: Pricing and rate limits
- Meta Model API documentation: Choose a model
Image: header artwork from Meta’s Muse Ember announcement, used for editorial purposes. FreeAnswers is not affiliated with Meta.