🔍 Read the full analysis: Muse Wonder Full’s AI Frontier Move Meets Muse Wonder’s Charge on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Artificial Analysis’ October 12 evaluation gave Meta’s Muse Wonder Full an Intelligence Index score of 56, while the API-available Muse Wonder scored 52. The API model’s estimated cost per index task was $4.52, compared with $0.72 for GPT-6.1 Sol at the same score; Muse Wonder Full is only available through Muse Code, with no public per-token price.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that scored 52 and 56, respectively, on the Artificial Analysis Intelligence Index v4.3, published October 12. The results put the stronger model near the top of the benchmark, but the version customers can use through Meta’s API costs an estimated $4.52 per index task—the same score as GPT-6.1 Sol, which the evaluation priced at $0.72 per task.
Artificial Analysis said Muse Wonder Full scored 56, eight points above Meta’s Muse Spark 1.3 score of 48, and improved on all ten evaluations in the index. Among 27 configurations compared in the source report, only Claude Opus 5.5 at 58 scored higher; Claude Sonnet 5.5 also scored 56. These are benchmark results, not evidence that every user or task will see the same performance.
The two models differ in access. Muse Wonder is available through Meta’s API and inside Muse Code, Meta’s coding agent. Its list prices are $4 per million input tokens, $0.20 for cached input tokens and $20 per million output tokens. Muse Wonder Full is available only to Muse Code users; Meta has not published a per-token price for it, so its cost per task cannot be calculated from the information provided.
Both models are described as having a one-million-token context window, accepting text, images, audio and video, and returning text. Artificial Analysis reported Muse Wonder Full at 57% on Terminal-Bench 4.0, versus 33% for Muse Spark 1.3. It also recorded a 30 score on AA-Omniscience for Full and 25 for Wonder, while reporting hallucination rates of 23% and 27%, respectively. Muse Spark 1.3’s reported hallucination rate was 33%.
Infographic 01 · Where Meta’s new pair lands
Muse Wonder Full ties for second. Muse Wonder ties GPT-6.1 Sol.
Artificial Analysis scores 27 frontier and open-weight configurations on ten evaluations. Meta’s two new models land near the top of the list. Six weeks earlier, Muse Spark 1.3 scored 48.
Intelligence Index score, 27 configurations
Higher is better. Reasoning effort in brackets, as Artificial Analysis tested it.
Availability: Muse Wonder runs on Meta’s API and in Muse Code. Muse Wonder Full is only available to Muse Code users and has no per-token price.
Source: Artificial Analysis, Muse Wonder analysis (12 Oct 2026) and model leaderboard, read 13 Oct 2026. Intelligence Index v4.3, ten evaluations.
Powered by Thorsten Meyer AI
Strong Scores, Higher API Costs
The release presents two different propositions for buyers: frontier-level benchmark performance in Muse Code and a separately purchasable API model with a lower score and a comparatively high estimated task cost. The strongest result cannot be bought by the token, while the accessible API model’s price may make it harder for developers to justify choosing it on cost-sensitive workloads.
Artificial Analysis estimated that a Muse Wonder task on its Intelligence Index costs $4.52. The same evaluation put GPT-6.1 Sol at $0.72 for a score of 52, and Gemini 4 Argon at $1.99 for a score of 53. Claude Sonnet 5.5 scored 56 at an estimated $5.46 per task. These figures are tied to the index’s task mix and evaluation method; they are not a universal price for typical API use.
The report also identifies input volume as a cost factor. Muse Wonder used 7.69 million input tokens per index task, compared with 6.03 million for Muse Spark 1.3. Artificial Analysis attributed $3.43, or 76%, of the $4.52 estimated task cost to input. That makes token consumption, alongside list prices, relevant to developers weighing the model for repeated or large-context work.
As an affiliate, we earn on qualifying purchases.
Meta’s Six-Week Model Step
The release follows Muse Spark 1.3 by about six weeks, according to the source report. In that short interval, the reported Intelligence Index score rose from 48 for Spark 1.3 to 56 for Muse Wonder Full, while the API model Muse Wonder reached 52. Artificial Analysis also reported improvements for Muse Wonder over Spark 1.3 on Terminal-Bench and AutomationBench-AA.
The reported Coding Agents Index compares models paired with their respective software harnesses, rather than models in isolation. Muse Code with Muse Wonder Full scored 68, matching Claude Code with Claude Sonnet 5.5 at 68; Claude Code with Opus 5.5 scored 66. Muse Code with Muse Wonder scored 63, and with Spark 1.3 scored 54. Those results describe complete model-and-tool configurations; they do not establish how Wonder Full would perform inside Claude Code.
The Intelligence Index comparisons also require care: a composite score summarizes multiple evaluations and does not settle performance for every task. Artificial Analysis reported Muse Wonder Full among the top three on six of ten evaluations, including a second-place GDPval-AA result of 1,857 Elo, close to Opus 5.5 at 1,866. The source report notes those results were within the models’ confidence intervals.
“The difference is the bill.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
Costs Beyond the Benchmark
Muse Wonder Full’s direct API price is not public in the supplied material because it is available only through Muse Code and has no per-token list price. The $4.52 estimate applies to Muse Wonder under Artificial Analysis’ index-task setup, not to Full. The source excerpt also ends during its cost breakdown, so it does not provide a complete accounting of all task-cost components.
The evaluation does not establish how the models will perform across the full range of production workloads, or whether the reported token use and costs will hold for other prompts, tools or usage patterns. Artificial Analysis’ benchmark findings are independent evaluations as described in the source, while broader claims about practical value remain dependent on each buyer’s tasks and verification needs.
Accuracy remains a consideration for consequential use. The reported hallucination rates were lower than Spark 1.3’s, but Muse Wonder Full’s 23% rate was above the 15% figure reported for Gemini 4 Argon among models scoring at least 45 on the index. Those benchmark rates do not predict the error rate in every application.
AI performance evaluation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
API Pricing and User Testing
Readers and developers can compare the published Muse Wonder API rates with their own token use and task requirements. Any decision about deployment should account for input volume, output volume, cached inputs where applicable, and the need to check model responses in workflows where errors carry a cost.
Further information would be needed to compare the full model’s economics with its performance: Meta has not provided a per-token price for Muse Wonder Full in the source material. More independent testing across real-world tasks could also show whether the benchmark advantages translate into consistent gains outside the evaluation suite.
For now, the confirmed picture is split: Muse Wonder Full has the stronger reported benchmark result but is confined to Muse Code, while Muse Wonder is the API option and carries a higher estimated index-task cost than several similarly scored rivals. Meta’s pricing and access decisions will determine how readily developers can test that trade-off.
As an affiliate, we earn on qualifying purchases.
Key Questions
What scores did the two Meta models receive?
Muse Wonder Full scored 56 and Muse Wonder scored 52 on Artificial Analysis’ Intelligence Index v4.3, published October 12, 2026.
Which model can developers access through Meta’s API?
Muse Wonder is available through Meta’s API and inside Muse Code. Muse Wonder Full is available only to Muse Code users, according to the supplied report.
How much does a Muse Wonder benchmark task cost?
Artificial Analysis estimated $4.52 per Intelligence Index task for Muse Wonder. The estimate reflects that evaluation’s task mix and is not a fixed cost for every API request.
Is Muse Wonder Full more expensive than Muse Wonder?
The supplied material does not provide a per-token price or task-cost estimate for Muse Wonder Full. It says the model is available only through Muse Code, so a direct price comparison is not possible from the reported information.
Does the benchmark prove the models are reliable for high-stakes work?
No. The reported scores and hallucination rates are benchmark results, not guarantees for every use. The source reports lower hallucination rates than Muse Spark 1.3 but says verification remains relevant when a wrong answer could have serious consequences.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
