🔍 Read the full analysis: Muse Ember By Meta: What A Score Of 36 Costs 14 Cents A Task on ThorstenMeyerAI.com
Get the little things that make your day delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Meta released Muse Ember on November 9, 2026. Artificial Analysis scored its max-effort version at 36, equal to Claude Haiku 5.5, while estimating an average cost of $0.14 per index task versus $0.41 for Haiku. The evaluation also found lower factual accuracy but a lower hallucination rate, alongside mixed results across workflow and coding tests.
Meta released Muse Ember on November 9, and an evaluation by Artificial Analysis gave its max-effort version an Intelligence Index score of 36 at an average cost of $0.14 per index task. The score matches Claude Haiku 5.5’s, while the same evaluation estimates Haiku costs $0.41 per task, making the price difference relevant to companies weighing AI models for repeated, high-volume work.
Artificial Analysis says the Intelligence Index combines nine evaluations. Its reported cost per task is an average API cost for completing one task in that index, not a guaranteed price for any particular business workload. At the higher “max” reasoning setting, Muse Ember scored 36; at “xhigh,” it scored 34 and cost $0.12 per index task. The max version produced 440 output tokens per second and took an average of 3.2 minutes per index task in the test.
Meta describes Muse Ember as the efficient tier of its Muse model line. It is proprietary, available through Meta’s API and the Muse Code coding agent, accepts text and images, and generates text. Artificial Analysis lists a one-million-token context window. Its listed input and output prices are $0.10 and $0.50 per million tokens, respectively, the same rates reported for Claude Haiku 5.5 and GPT-6 Luna. The evaluator attributes Muse Ember’s lower task cost relative to Haiku partly to cached input costing 60% less and partly to using about a third as many output tokens across the index.
Results varied by test. Muse Ember scored 68.1% on AutomationBench-AA, matching Meta’s more expensive Muse Wonder and exceeding Haiku 5.5’s 35.4% in the figures supplied by Artificial Analysis. It beat Haiku on Terminal-Bench 4.0, 40.9% to 32.8%, but scored lower on FrontierCode 2.0, 27.2% to 34.9%. It also scored 22.8% on the GDP.pdf document test, compared with Haiku’s 20.8%, while trailing on the AA-Briefcase knowledge-work test, where the reported ratings were 1,521 and 1,574 Elo.
Muse Ember by Meta What a score of 36 costs
Artificial Analysis matched Muse Ember’s max-effort score with Claude Haiku 5.5’s, while estimating a lower average benchmark-task cost. Results varied by test, so businesses should validate performance and cost on their own workflows.
Cost versus score
The reported index scores match, but the benchmark task costs differ.
| Model | Index score | Cost per task |
|---|---|---|
| Muse Ember, max effort | 36 | $0.14 |
| Claude Haiku 5.5 | 36 | $0.41 |
| Muse Ember, xhigh | 34 | $0.12 |
Mixed results across tests
Muse Ember led Haiku on some supplied evaluations and trailed on others.
AutomationBench-AA
Muse Ember scored 68.1%, matching Muse Wonder and exceeding Haiku 5.5’s 35.4%.
Different results by test
It led on Terminal-Bench 4.0, 40.9% to 32.8%, but trailed on FrontierCode 2.0, 27.2% to 34.9%.
A close document result
On GDP.pdf, Muse Ember scored 22.8% versus Haiku’s 20.8%; it trailed on AA-Briefcase, 1,521 to 1,574 Elo.
Factuality and refusals
The cited AA-Omniscience evaluation shows a trade-off, not a guarantee for real-world use.
What the cost means for teams
The benchmark price can inform evaluation, but it is not a forecast of a company’s bill.
Compare accuracy, refusals, latency and total cost on your prompts and workflows.
Prompt length, cached context, output volume and model settings can change costs.
Decide when to route unanswered requests to another model or a human reviewer.
Lower Costs for Repeated AI Tasks
The reported price difference could matter most for products that send models large volumes of routine requests, such as tagging, extraction, form completion or repeated agent steps. In Artificial Analysis’s test, the average cost for one index task with Muse Ember was about one-third of Haiku 5.5’s. That is a benchmark comparison, not a forecast of savings for a company: the actual bill depends on its prompts, context, usage and task mix.
The results also point to a practical trade-off rather than a simple ranking. Muse Ember’s lower cost comes alongside weaker performance on the cited factual-knowledge test, while its tendency to decline questions may reduce unsupported answers in some settings. Businesses would need to test the model on their own tasks and decide how to handle refusals, including whether to route unanswered requests to another model or a human reviewer.
Products Worth Considering
LEVOIT Vital 200S-P Smart Air Purifier for Large Rooms up to 1875 ft², Quiet 27 dB Operation, AHAM VERIFIDE Certified, Air Quality Monitor,
Garmin Forerunner 165 GPS Running Smartwatch, 43 mm, Whitestone
Garmin Forerunner 265S 42 mm GPS Running Watch, Light Pink and Powder Gray
As an affiliate, we earn on qualifying purchases.
How the Benchmark Compares Models
Artificial Analysis’s figures compare models under specified test conditions and reasoning settings. The Intelligence Index score is not a general guarantee of performance, nor does one score show how a model will behave in every deployment. The reported $0.14 is the average cost of completing an index task using Muse Ember at max effort; it should not be read as a fixed charge for any user request.
The source’s comparison places Muse Ember’s score of 36 alongside Haiku 5.5 at 36, GPT-6 Luna at 28 and Muse Wonder at 44. It reports Muse Wonder’s index-task cost as $6.95. The source also cites Claude Fable 5.2 scoring 65 at its highest effort level for $26.54 per task. Those comparisons illustrate that higher index scores can come with higher tested costs, but they do not establish that a more expensive model is better for every application.
Limits of the Reported Results
Artificial Analysis’s figures do not establish how Muse Ember will perform on a particular company’s data or workflow. The source material does not provide independent replication of the test results, a full breakdown of every evaluation’s weighting, or enough information to calculate the cost of a specific deployment. Benchmark costs and scores can differ from real-world results as prompt lengths, cached context, output volume and model settings change.
On AA-Omniscience, a test that allows models to abstain, Artificial Analysis reports Muse Ember answered 21.1% of questions correctly and declined 64%. Its hallucination rate was 17.4%, compared with 34.2% for Haiku 5.5 and 79.3% for GPT-6 Luna in the cited data. These are results from that evaluation; the source does not establish that the same rates apply across other topics or real-world use. It also remains unclear how Meta’s model will perform in broader independent testing or how users will experience its refusals in production.
Testing Before Wider Deployment
The next practical step for organizations interested in Muse Ember is to compare it with alternatives on their own tasks, measuring accuracy, refusal rates, latency and total cost. For work relying on supplied documents or retrieved information, teams can test whether that added context improves results and whether the lower cache price changes the overall economics. They can also decide how a workflow should respond when the model declines an answer.
Artificial Analysis’s published data provides a starting point, but the supplied source does not announce a further evaluation date or a new Meta release milestone. Broader independent testing and real deployment results will help clarify whether Muse Ember’s benchmark cost advantage carries over to specific applications.
Key Questions
What did Artificial Analysis report about Muse Ember?
Artificial Analysis reported an Intelligence Index score of 36 for Muse Ember at max effort and an average cost of $0.14 per task in that index. Its xhigh version scored 34 and cost $0.12 per index task.
How does Muse Ember’s reported task cost compare with Haiku 5.5?
In the cited index results, Muse Ember at max effort cost $0.14 per task, while Claude Haiku 5.5 at max effort cost $0.41. Both scored 36. The figures are benchmark averages, not guaranteed prices for other workloads.
Does the evaluation show Muse Ember is better at coding?
No single conclusion follows across the coding tests cited. Muse Ember scored 40.9% on Terminal-Bench 4.0, ahead of Haiku 5.5’s 32.8%, but scored 27.2% on FrontierCode 2.0, below Haiku’s 34.9%.
What does the evaluation say about Muse Ember’s factual answers?
On Artificial Analysis’s AA-Omniscience test, Muse Ember answered 21.1% of questions correctly and declined 64%. Its reported hallucination rate was 17.4%. These are results from that test and should not be assumed to predict performance on every factual task.
Source: ThorstenMeyerAI.com
