TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Artificial Analysis gave Meta’s Muse Ember a score of 36 on its Intelligence Index, matching Claude Haiku 5.5 while estimating a lower average cost per index task. The same evaluation found lower factual accuracy for Muse Ember, alongside frequent refusals and a lower hallucination rate than several compared models.
Meta released Muse Ember on November 9, 2026, and an Artificial Analysis evaluation gave its max-effort version an Intelligence Index score of 36, matching Claude Haiku 5.5 in the same index. Artificial Analysis estimates an average cost of $0.14 per index task for Muse Ember, compared with $0.41 for Haiku 5.5; its tests also found lower factual accuracy but a lower hallucination rate for Ember.
Artificial Analysis describes the Intelligence Index as a collection of nine evaluations. In its model data accessed November 9 and 10, Muse Ember at max reasoning effort scored 36, while the xhigh version scored 34. The max version averaged 3.2 minutes per index task through Meta’s API and generated 115 million output tokens across the index. The xhigh version averaged 2.8 minutes and generated 90 million output tokens, at an estimated $0.12 per task.
The reported price per task is distinct from token list prices. Muse Ember’s listed rates are $0.10 per million input tokens and $0.50 per million output tokens; cached input is listed at $0.004 per million tokens. Artificial Analysis attributes the lower task estimate relative to Haiku 5.5 to cheaper cached input and lower output-token use in its agentic tests. These are benchmark cost estimates, not a guarantee of what a customer will pay for a different workload.
Results varied by task. Muse Ember scored 68.1% on AutomationBench-AA, matching Meta’s more expensive Muse Wonder in that evaluation and exceeding Haiku 5.5’s 35.4%. It scored 22.8% on the GDP.pdf document test, compared with Haiku’s 20.8%, but trailed on the AA-Briefcase knowledge-work test by Elo rating, 1,521 to 1,574. Coding results also split: Ember led on Terminal-Bench 4.0, 40.9% to 32.8%, while Haiku led on FrontierCode 2.0, 34.9% to 27.2%.
Muse Ember’s Score and “I Don’t Know”
Artificial Analysis found Muse Ember matched Claude Haiku 5.5 on its Intelligence Index, with a lower estimated task cost and frequent refusals on a factual test.
A lower estimated task cost
The index covers nine evaluations. In this comparison, Ember matched Haiku 5.5’s score at a lower estimated cost.
| Model | Index score | Cost per task |
|---|---|---|
| Muse Ember, max | 36 | $0.14 |
| Claude Haiku 5.5 | 36 | $0.41 |
Performance varies by task
Ember led on some workflow and coding tests, while Haiku led on others.
AutomationBench-AA
Ember scored 68.1%, above Haiku 5.5’s 35.4%.
Split results
Ember led Terminal-Bench 4.0; Haiku led FrontierCode 2.0.
Frequent refusals
On AA-Omniscience, Ember declined 64% of questions and answered 21.1% correctly.
What the findings mean
The benchmark points to a trade-off: less guessing, but more unanswered factual questions.
- Repeated, structured tasks may merit a cost comparison.
- Lower hallucination in this test may matter where unsupported answers carry a cost.
- Refusals can interrupt work that needs factual answers.
- Actual cost depends on task design, token use, caching and API pricing.
Test on representative work
Benchmark results do not establish how Ember will perform in a company’s own systems.
Compare Ember with alternatives on representative internal tasks.
Measure completion quality, refusals, human review time, token use and total cost.
Check obscure and out-of-scope questions if factual answers matter.
Where Lower Cost Could Matter
The reported price gap could matter for teams that run high volumes of repeated, structured tasks such as routing, extraction, tagging and form filling. If those jobs resemble the index’s agentic tests, the lower estimated cost per task may make Muse Ember worth evaluating alongside other models. The published figures alone do not establish savings for a particular company: task design, token use, caching, reliability requirements and API pricing all affect actual costs.
Its refusal pattern also affects where it may fit. Artificial Analysis found that Muse Ember often declined questions in its factual-knowledge test. That may be useful in settings where an unsupported answer carries a cost, but refusals can interrupt work that depends on a model supplying a factual response. Businesses would need to assess both outcomes on their own examples, including how often a refusal requires human follow-up.
The benchmark does not show that one model is best for every use. Muse Ember’s stronger results on some workflow and coding tests sit alongside weaker results on another coding benchmark and a factual test. The relevant comparison depends on the job a team needs to complete, not just the headline index score.
Products Worth Considering
LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz
LG UltraGear 25G590B-B 25-Inch Full HD 1000Hz Gaming Monitor
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Twilight Blue
As an affiliate, we earn on qualifying purchases.
How the Benchmark Compares
Muse Ember is the efficient tier in Meta’s Muse model line. The source describes it as a proprietary model available through Meta’s API and the Muse Code coding agent. It accepts text and images as input and produces text. Artificial Analysis tested two reasoning-effort settings, max and xhigh; comparisons in the source generally use the max setting unless otherwise specified.
At max effort, Artificial Analysis reported the same index score for Muse Ember and Claude Haiku 5.5, but different estimated task costs: $0.14 for Ember and $0.41 for Haiku. The comparison is limited to the index methodology and the model configurations listed. The source also reports that Muse Wonder scored 44 at an estimated $6.95 per task, while Claude Fable 5.2 at its highest effort scored 65 at $26.54 per task. Those comparisons illustrate the cost differences in this dataset; they do not measure the value of each model for every use.
One result that needs separate context is the Artificial Analysis Coding Agent Index. In Meta’s Muse Code, Ember scored 46, compared with 37 for Haiku 5.5 in Claude Code and 31 for GPT-6 Luna in Codex. These are results for models used through different coding-agent products, so the scores describe those evaluated setups rather than a direct comparison of the underlying models in identical conditions.
Limits of the Current Results
The available source material reports benchmark results, not independent evidence about performance in customer deployments. It does not provide results for every real-world workload, nor does it establish how often users will receive a refusal, an incorrect answer or a successful completion in their own systems. Production accuracy, refusal rates and total operating costs remain dependent on the task and implementation.
On AA-Omniscience, Artificial Analysis reported that Muse Ember answered 21.1% of questions correctly and declined 64%. Its reported hallucination rate was 17.4%, compared with 34.2% for Haiku 5.5 and 79.3% for GPT-6 Luna. The source characterizes this as a knowledge weakness paired with less guessing. Those figures describe that specific test; they should not be treated as universal rates across topics or use cases.
The source material does not specify how the benchmark results will change with future model updates, or whether Meta will revise prices or availability. It also does not include enough detail to independently reproduce the reported task-cost estimates. Readers should treat those figures as Artificial Analysis’s measurements and estimates, dated November 9 and 10, 2026.
Testing Ember on Real Work
The next practical step for organizations considering Muse Ember is to compare it with alternatives on representative internal tasks. A useful evaluation would track completion quality, refusals, human review time, token use and total cost, rather than relying only on the Intelligence Index score. Teams that need factual answers should include obscure or out-of-scope questions in those checks, since the benchmark suggests Ember may decline often.
Meta lists the model through its API and Muse Code, according to the source. Further benchmark results, product updates or customer deployment data may clarify where Ember performs well outside the evaluations cited here. Until then, the reported findings support a narrower conclusion: Muse Ember matched Haiku 5.5’s index score at a lower estimated task cost in this test, while showing a distinct trade-off in factual answering and refusal behavior.
Key Questions
What score did Muse Ember receive?
Artificial Analysis gave Muse Ember at max effort a score of 36 on its Intelligence Index, which covers nine evaluations. The xhigh version scored 34 in the model data accessed November 9 and 10, 2026.
How much does a Muse Ember index task cost?
Artificial Analysis estimated an average of $0.14 per index task for Muse Ember at max effort. That is a benchmark estimate, not a quoted price for every customer workload. The source lists token prices separately.
Why does Muse Ember often say it does not know?
In Artificial Analysis’s AA-Omniscience test, Muse Ember declined 64% of questions and answered 21.1% correctly. The evaluation reported a 17.4% hallucination rate. These results apply to that test and do not establish refusal or accuracy rates for other tasks.
Is Muse Ember better than Claude Haiku 5.5?
The reported results do not establish an overall winner. Both scored 36 on the Intelligence Index, while Muse Ember had a lower estimated task cost in that evaluation. Their results differed across workflow, document, coding and factual-knowledge tests.
Source: Thorsten Meyer AI
