TL;DR
Get everyday essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Meta released Muse Ember on November 9, 2026. Artificial Analysis scores its max-effort version at 36 on its Intelligence Index, matching Claude Haiku 5.5 at a lower estimated cost per index task, but reports low factual accuracy and frequent refusals on a knowledge test.
Meta released Muse Ember on November 9, and an Artificial Analysis evaluation scores its max-effort version at 36 on the Intelligence Index, level with Claude Haiku 5.5. Artificial Analysis estimates an index task costs $0.14 with Muse Ember, compared with $0.41 for Haiku 5.5, while also finding that Ember often declines difficult factual questions rather than answering them.
The figures come from Artificial Analysis, which says its Intelligence Index combines nine evaluations. The service lists Muse Ember as a proprietary model available through Meta’s API and the Muse Code coding agent. It accepts text and images and produces text. Artificial Analysis tested two reasoning settings: max, which scored 36, and xhigh, which scored 34. The xhigh run was estimated at $0.12 per index task and took 2.8 minutes, compared with 3.2 minutes for max.
Artificial Analysis lists token prices of $0.10 per million input tokens and $0.50 per million output tokens. It says the estimated task cost reflects how much of each model’s input is cached and how many output tokens it uses: Ember’s cached input rate is $0.004, which the source describes as 60% below Haiku 5.5’s, and Ember used roughly a third as many output tokens as Haiku across the index. The same source reports 440 output tokens per second for Ember at max and a one-million-token context window.
Results vary by test. Ember scored 68.1% on AutomationBench-AA, matching Meta’s more expensive Muse Wonder and exceeding Haiku 5.5’s 35.4% in the cited results. On coding evaluations, Ember scored 40.9% on Terminal-Bench 4.0 against Haiku’s 32.8%, but 27.2% on FrontierCode 2.0 against Haiku’s 34.9%. In Artificial Analysis’s Coding Agent Index, Ember in Meta’s Muse Code scored 46, compared with 37 for Haiku in Claude Code and 31 for GPT-6 Luna in Codex.
Muse Ember: lower task cost with a trade-off in answers
Artificial Analysis scores Muse Ember level with Claude Haiku 5.5 on its Intelligence Index, but reports frequent refusals on a factual knowledge test.
A cheaper benchmark run
These costs reflect Artificial Analysis’s evaluation setup, not every company’s workload.
- Index score: 36
- Estimated task cost: $0.14
- Index score: 36
- Estimated task cost: $0.41
Factual answers are a weak spot
On AA-Omniscience, Ember often declined difficult questions.
Performance varies by task
Some results favor Ember; others favor Haiku 5.5.
AutomationBench-AA
Ember scored 68.1%, compared with Haiku 5.5 at 35.4%.
Mixed results
Ember led on Terminal-Bench 4.0, while Haiku 5.5 scored higher on FrontierCode 2.0.
What teams should check
Evaluate the model against real work before relying on benchmark estimates.
- Test representative tasks and difficult edge cases.
- Track accuracy, refusals, latency and API spending.
- Compare results with the model already in use.
A Lower Cost for Some Workloads
The benchmark points to a possible cost advantage for teams running repeated agent tasks, especially when those tasks reuse long inputs or produce relatively short outputs. Artificial Analysis estimates that Ember’s index work costs about a third as much per task as Haiku 5.5’s under its test conditions. That difference could matter at high volume, though the index estimate is not a price quote for a company’s own workflow.
The results also show why a single overall score can hide practical trade-offs. Ember’s workflow score is strong in the cited comparison, while its FrontierCode result trails Haiku’s. A business choosing a model would need to compare performance, refusals and cost on its own tasks. The benchmark alone does not establish that Ember is a suitable replacement across all uses.
Popular right now
Breville Barista Express Espresso Machine with Built-In Grinder, Black Sesame
Ninja Foodi 10-Qt DualZone Smart XL Air Fryer, 6-in-1, DZ550
Samsung 43-Inch M70H Series 4K Mini LED Smart TV (2026)
As an affiliate, we earn on qualifying purchases.
How Ember Compares on Cost
Artificial Analysis places Muse Ember at 36 and Haiku 5.5 at 36 on the Intelligence Index, with estimated task costs of $0.14 and $0.41 respectively. It also lists GPT-6 Luna at 28 for $0.15 per task and Meta’s Muse Wonder at 44 for $6.95. The source’s comparison describes the index task cost as the average API cost of one task in its evaluation, not the cost of every prompt or a standardized unit of business work.
The gap is larger at the top end of the cited results. Artificial Analysis lists Muse Wonder at eight index points above Ember and about 50 times its estimated task cost. It reports Claude Fable 5.2 at its highest effort level scoring 65 for $26.54 per task. Those comparisons show the price attached to these particular benchmark scores; they do not establish that the models perform equally on any given application.
Accuracy Depends on the Task
Artificial Analysis reports that Muse Ember answered 21.1% of AA-Omniscience questions correctly and declined 64%. Its reported hallucination rate was 17.4%. In the same comparison, Haiku 5.5 had 34.7% accuracy and a 34.2% hallucination rate; GPT-6 Luna’s reported hallucination rate was 79.3%. These figures describe one benchmark and do not establish how often each model will refuse or make mistakes in other settings.
The available source material does not provide independent replication of these results, full evaluation methods, or performance on a specific organization’s data. It also does not supply enough detail to judge how much the refusal behavior affects practical use across different question types. The benchmark’s task costs and scores should be treated as estimates tied to Artificial Analysis’s setup, rather than guaranteed outcomes.
Test Ember on Real Work
The immediate next step for prospective users is to evaluate Muse Ember on representative tasks, including edge cases where a confident wrong answer would carry a cost. Teams can track accuracy, refusal rates, latency and API spending alongside task completion, then compare those results with their current model. The source advises measuring performance on a business’s own tasks because the index may not reflect its workload.
Meta’s model is available through its API and Muse Code, according to Artificial Analysis. The source material does not give a date for further benchmark updates or identify a planned next release. More evidence from independent evaluations and real deployments would help establish how consistently Ember’s cost and refusal profile holds beyond the reported tests.
Key Questions
What score did Muse Ember receive?
Artificial Analysis gave Muse Ember at max reasoning effort a score of 36 on its Intelligence Index, which combines nine evaluations. Its xhigh setting scored 34.How much does a Muse Ember index task cost?
Artificial Analysis estimates an average cost of $0.14 per index task for Ember at max effort and $0.12 at xhigh. These estimates describe the source’s evaluation, not a universal cost for every task.Why does Muse Ember often say it does not know?
On Artificial Analysis’s AA-Omniscience test, which allows a model to decline questions, Ember refused 64% of questions. The source reports 21.1% accuracy and a 17.4% hallucination rate in that test.Is Muse Ember better than Claude Haiku 5.5?
The benchmark results are mixed. Both scored 36 on the Intelligence Index, while Ember’s estimated index task cost was lower. Ember did better on some cited workflow and coding tests, but Haiku scored higher on FrontierCode 2.0 and AA-Omniscience accuracy.Source: Thorsten Meyer AI
