Meta
Muse Ember
Meta's smallest Muse model, built for the steps an agent repeats all day: tool calls, coding subtasks, routing, extraction and long documents.
Run this model
meta/muse-ember
Use an endpoint supported by this model.
Get a key for this setup →const response = await fetch('https://api.minirouter.sh/v1/chat/completions', { method: 'POST', headers: { Authorization: `Bearer ${process.env.MINIROUTER_KEY}`, 'Content-Type': 'application/json', }, body: JSON.stringify({ model: 'meta/muse-ember', messages: [{ role: 'user', content: 'Why is the sky blue?' }], max_tokens: 1024, }),}) const data = await response.json()console.log(data.choices[0].message.content)Reasoning defaults to max and uses output tokens. Set reasoning effort and max_tokens explicitly to control cost per request.
Model overview
- Input / 1M
- $0.105
- All-in MiniRouter rate
- Output / 1M
- $0.525
- All-in MiniRouter rate
- Context
- 1M
- 1,048,576 tokens
- Max output
- 1M
- tokens
- Reasoning
- minimal → max
- always on; defaults to max
- Cost per task
- $0.14
- Artificial Analysis Index v5; Spark 1.3 is $2.39
- Announced
- 9 Nov 2026
- publisher release
Compare text-model alternatives or estimate token costs for your workload.
Near Spark 1.3's score at a fraction of the cost.
Meta reports Muse Ember one point behind Muse Spark 1.3 on the Artificial Analysis Index, at 6% of the cost per task. Against Claude Haiku 5.5 and GPT-6 Luna it leads or ties nine of ten rows; Haiku 5.5 is ahead on FrontierCode. MiniRouter plotted the table and did not run the evaluations.
Artificial Analysis Index v5
$0.14 per task, against $2.39 for Spark 1.3.
Cost to run one Intelligence Index task at list prices, both models at max reasoning effort.
- Muse Ember
- 36Index score
- Muse Spark 1.3
- 37Index score
- Cost per task
- 6%of Spark 1.3
Cost per Index task, USD
$0 to $2.50
- Muse Ember0.14
- Muse Spark 1.32.39
View chart data
| Measure | Muse Ember | Muse Spark 1.3 |
|---|---|---|
| Intelligence Index | 36 | 37 |
| Cost per Index task | $0.14 | $2.39 |
| Input / 1M | $0.10 | $1.25 |
| Output / 1M | $0.50 | $4.25 |
Ten evaluations
Agents, coding, multimodal work
- Muse Ember
- Comparison models
Agents
AutomationBench-AA
- Muse Ember37.8%
- Claude Haiku 5.513.9%
- GPT-6 Luna20.2%
0 to 100%
Agents
Agents' Last Exam
- Muse Ember61.6%
- Claude Haiku 5.557.0%
- GPT-6 Luna57.1%
0 to 100%
Agents
Personal Superintelligence Index
- Muse Ember46.8%1
- Claude Haiku 5.536.1%
- GPT-6 Luna37.0%
0 to 100%
Coding
Terminal-Bench 4.0
- Muse Ember40.9%2
- Claude Haiku 5.539.2%
- GPT-6 Luna16.4%
0 to 100%
Coding
Coding Agent Index-AA
- Muse Ember46
- Claude Haiku 5.537
- GPT-6 Luna31
0 to 100
Coding
FrontierCode 2.0-AA
- Muse Ember27.2%
- Claude Haiku 5.534.9%
- GPT-6 Luna25.2%
0 to 100%
Multimodal
GDP.pdf-AA
- Muse Ember22.8%
- Claude Haiku 5.520.8%
- GPT-6 Luna22.8%
0 to 100%
Multimodal
OSWorld 2.1 · partial
- Muse Ember69.5%
- Claude Haiku 5.568.9%
- GPT-6 Luna50.5%
0 to 100%
Multimodal
PerceptionBench · with tools
- Muse Ember80.6%
- Claude Haiku 5.578.6%
- GPT-6 Luna62.8%
0 to 100%
General
Artificial Analysis Index v5
- Muse Ember36
- Claude Haiku 5.536
- GPT-6 Luna28
0 to 100
View chart data
| Benchmark | Muse Ember | Claude Haiku 5.5 | GPT-6 Luna |
|---|---|---|---|
| AutomationBench-AA | 37.8% | 13.9% | 20.2% |
| Agents' Last Exam | 61.6% | 57.0% | 57.1% |
| Personal Superintelligence Index | 46.8% | 36.1% | 37.0% |
| 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% |
| PerceptionBench · with tools | 80.6% | 78.6% | 62.8% |
| Artificial Analysis Index v5 | 36 | 36 | 28 |
One rate at any prompt length.
Meta list · per 1M tokens
- Input
- $0.10
- Cached input
- $0.004
- Output
- $0.50
View chart data
| Rate | USD per 1M tokens |
|---|---|
| Input | $0.10 |
| Cached input | $0.004 |
| Output | $0.50 |
Reasoning is always on.
- [ 01 ]
The default is max
A request without a reasoning setting runs at max effort. Set minimal or low for routing, extraction and tool calls.
- [ 02 ]
Reasoning cannot be turned off
Meta's API rejects effort "none" with a 400. Use minimal for the lightest setting.
- [ 03 ]
Budget the output
Reasoning and the answer share max_tokens. Leave room for both, and compare completed-task cost across effort levels.
One route, standard clients.
https://api.minirouter.sh/v1meta/muse-ember/v1/chat/completions/v1/messages/v1/responsesConfigured providers. Priority follows the live catalog order.
| Provider | Priority | Input / 1M | Output / 1M | Parameters | Access |
|---|---|---|---|---|---|
vercel | 1 | $0.105 | $0.525 | max_tokens, temperature, stop, tools, tool_choice, reasoning, include_reasoning | MiniRouter balance |