🔍 Read the full analysis: Muse Wonder Full Joins The Frontier. Muse Wonder Charges Like It Did Too. on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Meta has released Muse Wonder and Muse Wonder Full, two proprietary models that Artificial Analysis evaluated on October 12, 2026. Wonder Full scored 56 on the Intelligence Index, while the API-available Wonder scored 52 and cost $4.52 per evaluated task in the analysis—more than several models with similar or higher scores.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that received strong but different results in an evaluation published by Artificial Analysis on October 12. Wonder Full scored 56 on the firm’s Intelligence Index, while the API-available Wonder scored 52; Artificial Analysis estimated one index task with Wonder at $4.52, raising questions about its cost compared with similarly scoring competitors.
Artificial Analysis says Wonder Full improved on all ten evaluations in its Intelligence Index v4.3. Its score of 56 was eight points above Muse Spark 1.3’s 48. Among 27 model configurations compared in the source report, only Claude Opus 5.5 at 58 scored higher, while Claude Sonnet 5.5 also scored 56. These are benchmark results, not a guarantee of performance in every use case.
The two models are not equally accessible. Muse Wonder is available through Meta’s API and in Muse Code, Meta’s coding agent. Its listed rates are $4 per million input tokens and $20 per million output tokens, with cached input listed at $0.20 per million tokens. Muse Wonder Full is available only to Muse Code users and has no per-token price in the material provided, so its cost per task cannot be independently calculated from that information.
On Artificial Analysis’s task-cost comparison, Wonder scored 52 and cost an estimated $4.52 per Intelligence Index task. GPT-6.1 Sol also scored 52 at $0.72 per task; Gemini 4 Argon scored 53 at $1.99; and GPT-6 Astra scored 53 at $3.26. The report attributes much of Wonder’s estimated cost to input processing: it used 7.69 million input tokens per task, which the source says accounted for $3.43, or 76% of the total.
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, Uneven Access and Cost
The release gives Meta a stronger position in frontier-model comparisons, especially for coding and agentic tasks. Artificial Analysis recorded Wonder Full at 57% on Terminal-Bench 4.0, up from 33% for Muse Spark 1.3, and placed it among the top three on six of the index’s ten evaluations. The report also found that Muse Code paired with Wonder Full scored 68 in its Coding Agents Index, tying the listed Claude Code and Sonnet 5.5 configuration.
But the strongest model is not offered as a generally purchasable API model, while the API version has lower benchmark scores and a higher estimated task cost than several similarly scoring alternatives. That distinction matters to organizations weighing model quality against operating expense: a high overall score does not by itself establish that a model is economical for a particular workload. The reported token usage suggests that input volume is a key part of Wonder’s task bill, though actual costs will vary with usage and implementation.
The evaluation also reports hallucination rates of 23% for Wonder Full and 27% for Wonder, compared with 33% for Spark 1.3. That is improvement in the reported measure, but the figures do not remove the need to check outputs in workflows where incorrect answers could have meaningful consequences.
As an affiliate, we earn on qualifying purchases.
From Muse Spark to Two Models
The release follows Muse Spark 1.3 by about six weeks, according to the source report. In Artificial Analysis’s comparison, Spark scored 48 on the Intelligence Index. Wonder Full’s 56 and Wonder’s 52 indicate a substantial rise in that evaluation, although the two new models occupy different product roles and are not available through the same channels.
Both models are described as proprietary, with a one-million-token context window. They accept text, images, speech or audio, and video as input, and return text. Wonder’s listed API rates are higher than those reported for Spark 1.3: $4 per million input tokens and $20 per million output tokens, compared with Spark’s $1.25 and $4.25. Artificial Analysis also found Wonder used more input tokens per index task than Spark, while producing fewer output tokens.
Artificial Analysis’s Coding Agents Index evaluates combinations of models and software harnesses rather than models alone. Its reported scores were 68 for Muse Code with Wonder Full, 68 for Claude Code with Sonnet 5.5, 66 for Claude Code with Opus 5.5, 63 for Muse Code with Wonder, and 54 for Muse Code with Spark 1.3. The result supports a comparison between those tested configurations; it does not establish that Wonder Full would score the same in a different coding environment.
“The difference is the bill: one Intelligence Index task costs $4.52 on Muse Wonder and $0.72 on Sol.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
Limits of the Cost Comparison
The supplied material does not provide a per-token price or independently calculated task cost for Muse Wonder Full, which is restricted to Muse Code. It also does not specify enough about the task-cost methodology to determine how closely the index-task estimate would match a particular customer’s workload, prompts, cached inputs, or deployment setup.
The results are attributed to Artificial Analysis, but the source material does not include full methodology details, confidence intervals for every comparison, or independent replication. Benchmark scores describe performance on selected tests; they do not settle how the models will perform across all production tasks. The source also ends during the breakdown of Wonder’s input and output costs, so no further cost components can be confirmed from the material provided.
As an affiliate, we earn on qualifying purchases.
API Use and Further Evaluation
Meta’s next practical test is how Wonder performs and bills in real API use, particularly for tasks that process large amounts of input. Developers can compare their own token consumption and output quality against the listed rates, while treating Artificial Analysis’s $4.52 task estimate as a benchmark-specific comparison rather than a universal price.
Further details would help clarify Wonder Full’s access, pricing, and cost per task, as well as the methods behind the reported evaluations. Until then, the available evidence points to a strong benchmark result for the Muse family, with a clear separation between the higher-scoring Muse Code-only model and the more widely accessible API model.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Meta release?
Meta released Muse Wonder and Muse Wonder Full, proprietary models with a one-million-token context window and text, image, audio or speech, and video input support.
How did the models score?
Artificial Analysis gave Muse Wonder Full a score of 56 and Muse Wonder a score of 52 on its Intelligence Index v4.3. The results reflect that evaluation, not performance on every possible task.
Can customers access Muse Wonder Full through Meta’s API?
The source report says Muse Wonder Full is available only to Muse Code users. Muse Wonder is the model listed as available through Meta’s API and in Muse Code.
Why is the reported cost of Muse Wonder drawing attention?
Artificial Analysis estimated $4.52 per Intelligence Index task for Wonder, compared with $0.72 for GPT-6.1 Sol, which had the same score of 52 in that comparison. The estimate is tied to the evaluation’s task and token usage, not a guaranteed cost for every user.
Does the reported hallucination improvement mean Wonder is safe to use without review?
No. The report lists hallucination rates of 27% for Wonder and 23% for Wonder Full, down from 33% for Spark 1.3. Those benchmark figures do not establish reliability for a specific workflow; verification remains relevant when errors carry costs.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
