Muse Ember
- Provider
- Meta
- Status
- Current
- Context
- 1,000,000 tok
- Price
- $0.1 / $0.5 /MTok
Muse Ember is Meta’s low-cost model for the work agents repeat all day, such as tool calls, coding subtasks, routing and extracting data from long documents. It was released on 9 November 2026 on the Meta Model API as muse-ember-1.0 and in Muse Code, at $0.10/$0.50 per million tokens (Meta).
Artificial Analysis scored it 36 on its Intelligence Index v5.0.1 at max effort at launch, level with Claude Haiku 5.5, at $0.14 per completed index task against Haiku’s $0.41, and called it the cheapest model scoring 36. Meta published no SWE-bench figure and no parameter count, and the weights are not released.
Quick specs
| Muse Ember | |
|---|---|
| Released | 9 November 2026 |
| API model ID | muse-ember-1.0 |
| Input price | $0.10 / MTok ($0.30 above 200K tokens) |
| Cached input | $0.004 / MTok |
| Output price | $0.50 / MTok ($1.50 above 200K tokens) |
| Context window | 1,000,000 tokens |
| Max output | 512,000 tokens |
| Modalities | Text, image, audio and video in; text out |
| Knowledge cutoff | Not published |
| AA Intelligence Index | 36 at launch (v5.0.1, max); 34 at xhigh |
| Cost per AA index task | $0.14 |
| Output speed | 440 tokens a second (Artificial Analysis) |
| Weights | Not released |
Where it fits in the Muse family
Meta positions Muse Ember as the model agents call for routine, repeated steps, paired with Muse Wonder for hard problems and Muse Spark for coding and agentic work. It is available in the same places: the Meta Model API, Muse Code, and, Meta says, Google Cloud and Oracle Cloud.
Benchmark performance
The independent read (Artificial Analysis)
| Measure | Muse Ember at launch (9 Nov 2026) | Context |
|---|---|---|
| Intelligence Index v5.0.1 | 36 (max effort) | Claude Haiku 5.5 36, DeepSeek V4.1-Flash 33, GPT-6 Luna 28 |
| Index at xhigh | 34 | Haiku 5.5 at xhigh: 35 |
| Cost per index task | $0.14 | Haiku 5.5: $0.41; GPT-6 Luna: $0.15 |
| Output tokens, whole index | About 115 million | |
| Output speed | 440 tokens a second | |
| Time to first answer token | 4.7 seconds | Haiku 5.5: about 182 seconds |
Source: Artificial Analysis.
Meta’s launch table against Claude Haiku 5.5 (9 November 2026)
| Benchmark | Muse Ember | Claude Haiku 5.5 |
|---|---|---|
| AutomationBench (AA) | 37.8% | 13.9% |
| Coding Agent Index (AA) | 46 | 37 |
| Terminal-Bench 4.0 | 40.9% | 39.2% |
| FrontierCode 2.0 (AA) | 27.2% | 34.9% |
| OSWorld 2.1 | 69.5% | 68.9% |
Meta published these. The rows marked AA are Artificial Analysis runs; OSWorld 2.1 is Meta’s internal evaluation. Terminal-Bench 4.0 used a different harness for each model (mini-swe-agent for Muse Ember, Claude Code for Haiku 5.5), so that row is not a like-for-like comparison.
Pricing
| Muse Ember | Claude Haiku 5.5 | |
|---|---|---|
| Input | $0.10 | $0.10 (prompts to 100K) |
| Output | $0.50 | $0.50 (prompts to 100K) |
| Long prompts | $0.30 / $1.50 above 200K | $0.50 / $2.50 above 100K |
| Cost per AA index task | $0.14 | $0.41 |
The token prices match Haiku 5.5’s for short prompts. The difference is in how many tokens each model spends: on Artificial Analysis’s index, Muse Ember finished a task for about a third of Haiku 5.5’s cost.
Known limitations
No SWE-bench figure. Meta published no SWE-bench Pro or Verified result.
Closed weights. Unlike Muse Spark Open and Muse Glimmer, Muse Ember is API-only.
Meta’s comparisons are selective. Meta’s table mixes independent Artificial Analysis runs with its own evaluations, and one of its chart captions says Muse Ember’s costs were estimated at proposed pricing.
Frequently asked questions
When was Muse Ember released?
9 November 2026, on the Meta Model API as muse-ember-1.0 and in Muse Code.
How much does Muse Ember cost?
$0.10 per million input tokens and $0.50 per million output tokens, with cached input at $0.004. Prompts over 200,000 tokens cost $0.30 input and $1.50 output. Artificial Analysis measured $0.14 per completed Intelligence Index task at launch.
How does Muse Ember compare with Claude Haiku 5.5?
Both scored 36 on the Artificial Analysis Intelligence Index v5.0.1 at max effort when Muse Ember launched, and they charge the same $0.10/$0.50 for short prompts. Muse Ember used fewer tokens, so a completed index task cost $0.14 against Haiku 5.5’s $0.41. On Meta’s own table, Muse Ember leads on AutomationBench and the Coding Agent Index and trails on FrontierCode 2.0.
Are Muse Ember’s weights open?
No. Muse Ember is available only through Meta’s API, Muse Code and cloud partners.
What is Muse Ember’s context window?
1,000,000 tokens, with up to 512,000 output tokens. It takes text, image, audio and video input and returns text.