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 gives its max-effort version an Intelligence Index score of 36 and an average cost of $0.14 per index task, compared with $0.41 for Claude Haiku 5.5 at the same score. The benchmark also found lower factual accuracy for Muse Ember, alongside more frequent refusals and a lower hallucination rate.
Meta released Muse Ember on November 9, and Artificial Analysis’s Intelligence Index gives its max-effort version a score of 36, equal to Claude Haiku 5.5’s score. Artificial Analysis estimates an average cost of $0.14 per index task for Muse Ember, compared with $0.41 for Haiku, making the new model a lower-cost option on this benchmark while raising questions about its weaker factual knowledge.
Artificial Analysis tested Muse Ember at two reasoning effort settings. At max effort, it scored 36, with an average cost of $0.14 per index task, an output speed of 440 tokens per second and an average completion time of 3.2 minutes per task. At xhigh effort, it scored 34 and cost $0.12 per task, with a reported speed of 423 tokens per second and a 2.8-minute task time. These figures describe the index tests, not a typical customer workload.
The model’s listed prices are $0.10 per million input tokens and $0.50 per million output tokens, matching the listed rates for Haiku 5.5 and GPT-6 Luna. Artificial Analysis points to two reasons its estimated task cost is lower than Haiku’s: Muse Ember’s cached input costs $0.004, which the source says is 60% less than Haiku’s, and it used roughly one-third as many output tokens as Haiku to complete the index. Muse Ember has a one-million-token context window and accepts text and images as input.
Results varied across other tests. Muse Ember scored 68.1% on AutomationBench-AA, matching Meta’s more expensive Muse Wonder in that test; Haiku scored 35.4%. Muse Ember led Haiku on Terminal-Bench 4.0, 40.9% to 32.8%, but trailed it on FrontierCode 2.0, 27.2% to 34.9%. In Meta’s Muse Code tool, it scored 46 on Artificial Analysis’s Coding Agent Index; the source compares that with 37 for Haiku in Claude Code and 31 for GPT-6 Luna in Codex.
Muse Ember: equal score, lower benchmark cost
Released November 9, 2026, Muse Ember matched Claude Haiku 5.5 on Artificial Analysis’s Intelligence Index at a lower average task cost. Its factual accuracy and results across other tests complicate the comparison.
Same score, lower test cost
These are average costs on the index, not predictions for a company’s workload.
Performance varies by test
A matching overall score does not mean the models perform alike on every task.
| Evaluation | Muse Ember | Haiku 5.5 |
|---|---|---|
| AutomationBench-AA | 68.1% | 35.4% |
| Terminal-Bench 4.0 | 40.9% | 32.8% |
| FrontierCode 2.0 | 27.2% | 34.9% |
| AA-Omniscience accuracy | 21.1% | 34.7% |
Read factual results with refusals
On AA-Omniscience, the evaluation allowed models to decline questions.
- Answered 21.1% correctly
- Declined 64% of questions
- Reported hallucination rate: 17.4%
- Answered 34.7% correctly
- Reported hallucination rate: 34.2%
Test against real workflows
Teams weighing the lower benchmark cost can compare models using representative tasks and review criteria.
Include the tasks and context the business actually handles.
Measure accuracy and whether refusals suit the workflow.
Track latency and total API costs on the same workload.
Lower Costs, Uneven Results
The reported price difference could matter to businesses that run many model calls for routine tasks such as routing, extraction, tagging and form filling. At the benchmark’s average task costs, Muse Ember’s $0.14 is about one-third of Haiku 5.5’s $0.41. Actual savings would depend on the prompts, context, model settings and volume in a company’s own use; the index figures do not establish what those workloads would cost.
The comparison also shows why a single overall score may not settle a model choice. Muse Ember’s matching score does not mean it performs equally on every task. Its mixed coding results and low score on a factual knowledge test could affect whether it suits a particular workflow. A team that needs fast, repeated processing may weigh cost and speed differently from one that needs reliable answers to specialized factual questions.
Artificial Analysis’s AA-Omniscience test found Muse Ember answered 21.1% correctly and declined 64% of questions. Its reported hallucination rate was 17.4%, compared with 34.2% for Haiku 5.5. Those figures indicate a different response pattern on that test, not a guarantee that the model will abstain or avoid errors in every real-world exchange.
Products Worth Considering
Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB‑C Charging, Up to 2X More Active Noise Cancelling, Transparency Mode, Adaptive
Logitech MX Ergo S Advanced Wireless Trackball Mouse USB-C – Graphite
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
How the Benchmark Compares
Muse Ember is described by the source as the efficient tier of Meta’s Muse model line. It is proprietary and available through Meta’s API and in Muse Code, Meta’s coding agent. Artificial Analysis’s Intelligence Index combines nine evaluations; its task cost is an average API cost for a task in that index. The score and cost should be read together with the test results and their limits.
Among the comparisons listed by Artificial Analysis, Muse Wonder scored 44 for $6.95 per task, while Claude Fable 5.2 at its highest effort scored 65 for $26.54. The source describes Fable’s task cost as about 190 times Muse Ember’s, for 29 additional index points. These comparisons show the higher costs recorded for those models on the same index data, but do not establish that one model is better for every use.
On AA-Omniscience, Artificial Analysis compared Muse Ember’s 21.1% accuracy with 34.7% for Haiku 5.5 and 50.5% for Gemini 3.8 Flash at high effort. The evaluation allows models to decline questions. That design makes refusal rates relevant to interpreting accuracy and hallucination results: a model that answers fewer questions has fewer chances to make a wrong claim in the test.
Limits of the Current Evidence
The available figures are benchmark results published by Artificial Analysis, with the model pages accessed November 10, 2026. The source does not provide results from independent customer deployments or establish how the index performance translates to specific business tasks. It also does not state how performance changes across different prompts, integrations or production conditions.
It remains unclear how often Muse Ember will decline in ordinary use, whether its lower benchmark hallucination rate will hold across subject areas, and how its cost compares with alternatives on a company’s own traffic. The source says its token prices match those of Haiku 5.5 and GPT-6 Luna, but its task cost depends in part on cached input and output-token use during the index. Those usage patterns may differ elsewhere.
Testing in Business Workflows
The next useful comparison is likely to come from testing the model on real workloads. Organizations considering Muse Ember can compare its accuracy, refusals, latency and total API costs against other models using representative prompts and review criteria. For factual tasks, they can check whether the model’s willingness to abstain fits the workflow and how staff handle unanswered requests.
Meta’s model is currently described as available through its API and Muse Code. The source does not identify a future release date, additional benchmark schedule or planned change to pricing. Further results or deployment evidence could clarify whether the reported cost advantage and task-specific strengths carry over beyond Artificial Analysis’s evaluations.
Key Questions
What score did Muse Ember receive?
Artificial Analysis gave Muse Ember at max effort a score of 36 on its Intelligence Index. Its xhigh-effort version scored 34.
How much does Muse Ember cost per index task?
Artificial Analysis reports an average cost of $0.14 per task for Muse Ember at max effort and $0.12 at xhigh effort. These are index averages, not a quoted cost for every customer workload.
How does Muse Ember compare with Claude Haiku 5.5?
Both models scored 36 at max effort on the index. Artificial Analysis estimates $0.14 per Muse Ember task and $0.41 per Haiku task. Results differ across other tests, including coding and factual knowledge evaluations.
Does Muse Ember have lower hallucination rates?
On Artificial Analysis’s AA-Omniscience test, Muse Ember’s reported hallucination rate was 17.4%, compared with 34.2% for Haiku 5.5. The test also found Muse Ember declined 64% of questions; these benchmark results do not guarantee performance in other settings.
Source: Thorsten Meyer AI
