🔍 Read the full analysis: Muse Wonder Full’s AI Frontier Debut Comes With A Charge on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Artificial Analysis reported that Meta’s Muse Wonder Full scored 56 on its Intelligence Index, while the API-available Muse Wonder scored 52. The reported average cost for one index task with Muse Wonder was $4.52, compared with $0.72 for GPT-6.1 Sol, which also scored 52. Muse Wonder Full is available only through Muse Code and has no public per-token price.
Meta’s new Muse Wonder Full scored 56 on Artificial Analysis’ Intelligence Index, while the API model Muse Wonder scored 52; the evaluation also put Muse Wonder’s cost at $4.52 per index task. The October 12 assessment, cited in ThorstenMeyerAI.com’s October 13 report, describes a strong performance improvement over Muse Spark 1.3 but a high benchmark-task bill for the model customers can access through Meta’s API.
Artificial Analysis reported that Muse Wonder Full gained eight points over Muse Spark 1.3, which scored 48, and improved on all ten evaluations in the Intelligence Index v4.3. It placed among the top three on six evaluations. On Terminal-Bench 4.0, a test of work performed in a shell, Wonder Full scored 57%, up from 33% for Spark 1.3. The report also listed Wonder Full at 1,857 Elo on GDPval-AA, an evaluation of economically useful knowledge work, close to Claude Opus 5.5 at 1,866 and Claude Sonnet 5.5 at 1,838; the report says those results fall within each other’s confidence intervals.
The two models differ in access and price. Muse Wonder is available through Meta’s API and Muse Code, Meta’s coding agent, at list prices of $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Muse Wonder Full is available only in Muse Code; Meta has not provided a per-token price, so its cost per task cannot be calculated from public pricing.
For the Intelligence Index task-cost comparison, Artificial Analysis listed Muse Wonder at $4.52 per task. GPT-6.1 Sol scored the same 52 for $0.72, while Gemini 4 Argon scored 53 for $1.99. The report also said Muse Wonder used 7.69 million input tokens per task, compared with 6.03 million for Spark 1.3, and attributed 76% of Wonder’s task cost—$3.43—to input tokens.
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
Performance Gains Meet API Costs
The results matter because they separate Meta’s strongest showing from the model most customers can purchase directly. Wonder Full’s 56 score and coding-agent result suggest Meta can compete with leading configurations on the cited evaluations. In Artificial Analysis’ Coding Agents Index, Muse Code paired with Wonder Full scored 68, matching Claude Code with Sonnet 5.5. This is a result for those model-and-harness combinations, not proof that Wonder Full would score the same in another coding environment.
For buyers, the more immediate question is the cost of the API model. In the reported task comparison, Muse Wonder and GPT-6.1 Sol both scored 52, but their estimated costs were $4.52 and $0.72 respectively. Such comparisons can inform model selection, but they apply to the index’s task mix and token use; they do not establish the cost or value of every real-world workload. Wonder Full’s public per-token pricing is unavailable, leaving customers without an equivalent price comparison for Meta’s higher-scoring model.
As an affiliate, we earn on qualifying purchases.
Two Models, Different Access
The source report says Meta shipped Muse Wonder and Muse Wonder Full six weeks after Muse Spark 1.3. Both are proprietary models with a one-million-token context window. They accept text, images, speech or audio, and video as input, and return text. The evaluation and product details in this article are based on the figures reported by ThorstenMeyerAI.com and attributed there to Artificial Analysis.
The API model’s score is an improvement over its predecessor on several measures cited in the report: Terminal-Bench rose to 52% from Spark 1.3’s 33%, and AutomationBench-AA rose to 68% from 58%. The report also gives Muse Wonder an AA-Omniscience score of 25 and a 27% hallucination rate. For Wonder Full, it reports an AA-Omniscience score of 30 and a 23% hallucination rate, down from 33% for Spark 1.3. These metrics describe specific evaluations; they do not establish how often errors will occur in every use case.
“The difference is the bill.”
— Thorsten Meyer, ThorstenMeyerAI.com
AI model performance evaluation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Pricing and Real-World Performance
Meta’s public per-token price for Muse Wonder Full is not stated in the source material. Because access is limited to Muse Code, its cost per benchmark task cannot be compared with the API models on the same basis. The material also does not specify whether the reported task-cost estimates reflect typical customer workloads, or how costs would vary with different prompts, settings or usage patterns.
The reported scores come from independent evaluations, but benchmark results alone do not establish performance across all production tasks. Artificial Analysis’ cited hallucination figures also indicate that errors remain possible: the source lists 23% for Wonder Full and 27% for Wonder. The precise methods and uncertainty ranges for every score are not fully detailed in the supplied material, so broader claims about product reliability or value remain unverified here.
As an affiliate, we earn on qualifying purchases.
Watch API Adoption and Pricing
The next practical test will be whether developers adopt Muse Wonder through Meta’s API despite its reported task cost, and whether Meta changes or clarifies its pricing. Public per-token pricing for Wonder Full would allow buyers to compare the more highly rated model on a common basis, although its actual cost would still depend on use.
Readers should also watch for further evaluations and real-world reports that show how the models perform on specific workloads. Until then, the published evidence supports a narrower conclusion: Meta has posted higher benchmark scores with this release, while the API model’s reported cost and Wonder Full’s undisclosed task pricing leave important purchasing questions open.
As an affiliate, we earn on qualifying purchases.
Key Questions
Which Muse model can customers access through Meta’s API?
Muse Wonder is available through Meta’s API and inside Muse Code. Muse Wonder Full is reported as available only through Muse Code.
How did Muse Wonder Full score?
Artificial Analysis reported a score of 56 on the Intelligence Index v4.3, eight points above Muse Spark 1.3’s 48. The score summarizes performance on that index and is not a guarantee for every task.
How much does a Muse Wonder benchmark task cost?
The source report cites an Artificial Analysis estimate of $4.52 per Intelligence Index task. This is a benchmark-based comparison, not a fixed price for every customer workload.
What is the reported price of Muse Wonder Full?
The supplied material gives no per-token price for Muse Wonder Full. It says the model is available only through Muse Code, so its cost per task cannot be calculated from public API pricing in the report.
Does a high benchmark score mean the models do not hallucinate?
No. The report lists hallucination rates of 23% for Wonder Full and 27% for Wonder on the cited measure. Those figures do not predict every use case, but they indicate that verification may still be needed for consequential work.
Source: ThorstenMeyerAI.com
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
