🔍 Read the full analysis: What To Know About Meta’s Muse Ember’s Score And Per-Task Price on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Artificial Analysis reports that Meta’s Muse Ember (max) scored 36 on its Intelligence Index, matching Claude Haiku 5.5, while costing $0.14 per index task versus $0.41. The same tests show weaker factual accuracy for Ember, alongside more frequent refusals and a lower hallucination rate; results may not predict performance on every workload.
Meta’s Muse Ember, released November 9, scored 36 on Artificial Analysis’s Intelligence Index at its max reasoning setting, matching Claude Haiku 5.5 while costing $0.14 per index task versus $0.41, according to the testing firm’s data. The comparison makes Ember a lower-cost option on that benchmark, but the same results show a trade-off: it answered fewer difficult factual questions correctly and more often declined to answer.
Artificial Analysis says the Intelligence Index combines nine evaluations. Its listed “cost per task” is the average API cost of completing one task in that index, not a universal charge for every use of the model. The comparison is reported for both models at max reasoning effort. At xhigh effort, Ember scored 34 and its index-task cost was $0.12, with an average reported completion time of 2.8 minutes.
The per-token prices do not explain the full difference: the source lists Ember at $0.10 per million input tokens and $0.50 per million output tokens, the same as Haiku 5.5. Artificial Analysis attributes Ember’s lower task cost partly to cached input priced at $0.004, which it says is 60% cheaper than Haiku’s, and to Ember using roughly a third as many output tokens as Haiku across the index. Its reported output speed was 440 tokens per second, and the model has a one-million-token context window.
Results vary by test. Ember scored 68.1% on AutomationBench-AA, level with Meta’s more expensive Muse Wonder in the figures provided, and ahead of Haiku 5.5’s 35.4%. It also beat Haiku on Terminal-Bench 4.0, 40.9% to 32.8%, but trailed it on FrontierCode 2.0, 27.2% to 34.9%. On the AA-Omniscience factual test, Ember answered 21.1% correctly and declined 64% of questions; Artificial Analysis reported a 17.4% hallucination rate, compared with 34.2% for Haiku.
Meta’s Muse Ember score and task price
At max reasoning effort, Ember matched Claude Haiku 5.5 on Artificial Analysis’s Intelligence Index at a lower estimated cost per index task. Its factual accuracy and refusal results show trade-offs teams should test on their own workloads.
Lower benchmark cost, same score
The index combines nine evaluations; its task cost is an average for that test set, not a fixed price for every request.
- Score: 36 at max reasoning effort
- Average index-task cost: $0.14
- At xhigh effort: score 34 and $0.12 per task
- Score: 36 at max reasoning effort
- Average index-task cost: $0.41
- Same listed input and output token prices as Ember
Factuality has trade-offs
On AA-Omniscience, Ember was more likely to decline and had a lower reported hallucination rate, while answering fewer difficult factual questions correctly.
21.1% correct
Ember answered this test’s factual questions correctly at this rate.
64% declined
Frequent declines may need a human or another model to handle unanswered requests.
17.4% reported
The reported rate was lower than Haiku 5.5’s 34.2% on this evaluation.
Results vary by task
Ember led Haiku on some reported evaluations and trailed it on others.
| Evaluation | Muse Ember | Haiku 5.5 | Result |
|---|---|---|---|
| AutomationBench-AA | 68.1% | 35.4% | Ember higher |
| Terminal-Bench 4.0 | 40.9% | 32.8% | Ember higher |
| FrontierCode 2.0 | 27.2% | 34.9% | Ember lower |
| GDP.pdf | 22.8% | 20.8% | Ember higher |
Test it on real workflows
Benchmark costs may differ from a company’s actual costs, which depend on prompts, cached input, output volume, settings, refusals, and retries.
Compare Ember and alternatives with the same prompts and success criteria.
Measure accuracy, unsupported claims, refusals, completion time, token use, and fallback costs.
Account for human review or another model when Ember declines, and retest if pricing or the model changes.
Lower Cost, With a Factuality Trade-Off
The reported price gap could matter to organizations that run high volumes of repetitive model tasks, such as tagging, extraction, form filling, or steps in an automated workflow. At the index’s reported average, Ember’s task cost is about one-third of Haiku 5.5’s. That is a benchmark comparison, however, not proof that a business would see the same savings on its own prompts, data, or system design.
Ember’s refusal pattern also shapes where it may be useful. A model that often says it cannot answer may be preferable for workflows where unsupported factual guesses are costly, provided the system can route unanswered requests to a human or another model. That behavior can also interrupt workflows that need a direct answer. The lower hallucination rate and lower factual accuracy are measurements from a particular test, not guarantees about performance in production.
The model may fit tasks where a user supplies the relevant information, such as processing a document or retrieved records, better than tasks that depend on the model recalling obscure facts. Businesses would need to test accuracy, refusal rates, latency, and total costs on representative workloads before changing model routing.
Products Worth Considering
Logitech MX Ergo S Advanced Wireless Trackball Mouse USB-C – Graphite
Apple Mac mini Desktop Computer with M2 Pro, 16GB Memory, 512GB SSD, Gigabit Ethernet
Samsung 32 Odyssey QD-OLED G8 Gaming Monitor – 4K, 240Hz
As an affiliate, we earn on qualifying purchases.
Meta’s Efficient Muse Model
Muse Ember is described in the source as the efficient tier of Meta’s Muse model line, available through Meta’s API and in its Muse Code coding agent. It accepts text and images and generates text. Artificial Analysis tested it at max and xhigh reasoning effort; the figures cited here mainly compare max settings.
The index results place Ember alongside models with notably different scores and costs. Artificial Analysis lists Muse Wonder at 44 and $6.95 per index task, while Claude Fable 5.2 at its highest effort scored 65 at $26.54 per task. These are the source’s benchmark estimates, not direct measures of a company’s total cost or evidence that one model is best for every task.
Other reported results are mixed. Ember scored 22.8% on the GDP.pdf document test, compared with Haiku’s 20.8%, but trailed Haiku on the AA-Briefcase knowledge-work test, with scores of 1,521 and 1,574 Elo respectively. In the Artificial Analysis 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. These comparisons use different test setups and should be read as benchmark results, not universal rankings.
How Results Translate to Real Work
The source provides benchmark scores and estimated API costs, but it does not establish how Ember performs across customers’ specific workloads. The index-task figure is an average for tasks in one evaluation set; actual cost will depend on prompt length, cached input, output volume, model settings, and how a system handles refusals or retries.
The supplied material also does not give independent replication, detailed test methodology for each result, or a complete account of model availability, service limits, and pricing terms beyond the listed rates. The reported hallucination and accuracy figures should not be treated as guarantees. It remains unclear how often the refusal behavior would improve safety versus create extra work in day-to-day deployments.
Test Costs on Your Workflows
The practical next step for teams considering Ember is to compare it with alternatives on their own representative tasks, using the same prompts and success criteria. That evaluation should track correct answers, unsupported claims, refusals, completion time, token use, and the cost of any fallback model or human review.
Meta’s model is available through its API and Muse Code, according to the source. Further details on product terms, updates to pricing, and performance beyond the reported Artificial Analysis evaluations were not included in the material. Buyers should check current provider documentation and re-test if the model or pricing changes.
Key Questions
What score did Meta’s Muse Ember receive?
Artificial Analysis reported a score of 36 for Muse Ember at max reasoning effort on its nine-evaluation Intelligence Index. At xhigh effort, the reported score was 34.
How much does Muse Ember cost per index task?
Artificial Analysis reported an average of $0.14 per index task for Ember at max effort. This is an estimate for a task in its benchmark index, not a fixed price for every customer request.
How does Ember’s task cost compare with Claude Haiku 5.5?
Both scored 36 at max effort in the reported comparison. Artificial Analysis listed Ember at $0.14 per task and Haiku 5.5 at $0.41 per task.
Does Muse Ember answer factual questions reliably?
On Artificial Analysis’s AA-Omniscience test, Ember answered 21.1% correctly and declined 64% of questions. The test also recorded a 17.4% hallucination rate. Those results describe that evaluation and do not guarantee performance on other tasks.
What should companies test before using Muse Ember?
Teams should measure accuracy, refusals, hallucinations, speed, and total cost on their own workload. They should also account for what happens when Ember declines, including whether a human or another model must handle the request.
Source: ThorstenMeyerAI.com
