🔍 Read the full analysis: Muse Wonder Full Joins The Frontier. Muse Wonder Charges Like It Did Too. on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Meta has released Muse Wonder and Muse Wonder Full, with Artificial Analysis giving the Full model an Intelligence Index score of 56 and the API model a 52. The stronger version is limited to Meta’s Muse Code, while one evaluated task using the API model cost $4.52—far more than some similarly rated alternatives. The evaluation is independent, but the cost comparison reflects the benchmark’s task setup, not every customer’s usage.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that take text, images, audio and video as input. In an evaluation published Oct. 12 by Artificial Analysis, Wonder Full scored 56 on the Intelligence Index, while the version available through Meta’s API scored 52; the benchmark estimated a cost of $4.52 per task for the API model, making price a central question for customers comparing it with similarly rated systems.
Artificial Analysis’ Intelligence Index v4.3 puts Muse Wonder Full at 56, eight points above Muse Spark 1.3’s 48. The evaluator said Full improved on all 10 assessments in the index and ranked in the top three on six. Among 27 configurations cited in the source report, Claude Opus 5.5 in its max configuration scored 58, while Claude Sonnet 5.5 max also scored 56. These are benchmark results, not a guarantee of performance on every real-world task.
Muse Wonder scored 52, matching GPT-6.1 Sol max in the comparison. Meta makes Wonder available through its API and inside Muse Code, its coding agent. Its listed prices are $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Wonder Full is available only to Muse Code users, and Meta has not provided a per-token price for it, so its cost per task cannot be compared on the same basis.
On Artificial Analysis’ task-cost comparison, one Intelligence Index task using Wonder was estimated at $4.52. The report put GPT-6.1 Sol at $0.72 for a score of 52, Gemini 4 Argon at $1.99 for 53, and Claude Sonnet 5.5 at $5.46 for 56. Those figures depend on the evaluation’s token use and pricing assumptions; they are not fixed prices for all customer workloads.
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
The release places Meta in a stronger position in evaluations of coding and knowledge work, but access differs sharply between the two models. Wonder Full’s score of 56 is close to the leading systems listed in the report, and its coding-agent configuration tied the top result in Artificial Analysis’ separate Coding Agents Index. That index scored Muse Code with Wonder Full at 68, level with Claude Code using Sonnet 5.5.
For developers, the distinction is practical: the strongest result is available only through Muse Code, while the API-accessible Wonder scored four points lower. Businesses choosing an API model may weigh the benchmark score against task costs, latency, reliability and how well a model fits their own work. The reported $4.52 estimate raises a cost-efficiency issue for users running comparable evaluation tasks, but it does not by itself establish what any particular deployment will cost.
The report also records progress in areas relevant to agent use. Wonder Full scored 57% on Terminal-Bench 4.0, compared with 33% for Spark 1.3, and Wonder scored 52%. Artificial Analysis reported Wonder Full’s hallucination rate at 23%, down from 33% for Spark 1.3. Even with that improvement, the report’s figures show that errors remain a concern, especially for tasks where incorrect answers can carry financial or operational consequences.
As an affiliate, we earn on qualifying purchases.
Two Models, Different Access
Meta introduced the models about six weeks after Muse Spark 1.3, according to the source report. Both new models are proprietary, have a one-million-token context window and accept text, images, speech or audio, and video as input. Both return text. The key difference is where customers can use them: Wonder is offered through the API and Muse Code, while Wonder Full is restricted to Muse Code.
Artificial Analysis evaluates models across a set of tasks and combines results in its Intelligence Index. Its report also presents individual assessment scores, including GDPval-AA, which tests economically useful knowledge work. Wonder Full scored 1,857 Elo on GDPval-AA, compared with 1,866 for Opus 5.5 and 1,838 for Sonnet 5.5; the source says these results fall within the models’ confidence intervals. On the Coding Agents Index, the evaluator measures a model together with its software harness, so a score for Muse Code plus Wonder Full is not a direct comparison of the models alone.
The report says Wonder costs more per million tokens than Spark 1.3: $4 per million input tokens and $20 per million output tokens, versus Spark’s $1.25 and $4.25. Its task-cost estimate also reflects unusually high input use in the benchmark: 7.69 million input tokens for Wonder per index task, according to Artificial Analysis. That figure is specific to the evaluation and should not be treated as typical usage for every prompt or application.
“The bill is in the input, not the output.”
— Thorsten Meyer, ThorstenMeyerAI.com
Costs and Real-World Performance
The reported scores and costs come from a third-party evaluation, not from a disclosed Meta price for every possible workload. Wonder Full has no published per-token price in the source material, so its cost cannot be directly compared with API models. It is also unclear whether the $4.52 estimate for Wonder would apply to customers’ ordinary requests; the figure is tied to the benchmark task and its measured token use.
The source provides benchmark results but no details on Meta’s training methods, serving costs, independent replication or performance across specific customer applications. The relationship between benchmark scores and outcomes in production remains uncertain. Hallucination rates also vary by evaluation and task; the reported 23% for Wonder Full and 27% for Wonder should not be read as a universal error rate.
Customer Access and Independent Testing
Customers can assess Muse Wonder through Meta’s API and Muse Code, while Wonder Full remains available only through Muse Code, according to the source report. Meta has not stated when Full might become available as a separately priced API model or whether it plans to publish usage pricing for it.
Further independent evaluations and customer reports will help establish whether the benchmark gains hold across different tasks and deployment conditions. For buyers, the next useful evidence will include real workload costs, reliability results and clearer information about Wonder Full’s availability and pricing. Until then, the reported index scores offer one comparison point, while the task-cost figures should be interpreted within Artificial Analysis’ testing setup.
Key Questions
What are Muse Wonder and Muse Wonder Full?
They are two proprietary Meta models released in October 2026. Both accept text, images, audio and video as input and return text, according to the source report.
Which model can developers access through Meta’s API?
Muse Wonder is available through Meta’s API and in Muse Code. Muse Wonder Full is reported as available only to Muse Code users.
How did the models score in Artificial Analysis’ index?
Artificial Analysis gave Muse Wonder Full a score of 56 and Muse Wonder a score of 52 on Intelligence Index v4.3. These scores summarize performance across the evaluator’s tests and do not predict every application’s results.
What does the $4.52 cost figure mean?
It is Artificial Analysis’ estimate for one Intelligence Index task using Muse Wonder, based on the task’s measured token use and the model’s listed pricing. It is not a flat fee for every API request or a forecast of every customer’s bill.
Is the stronger model more expensive?
The source report says Meta has not published per-token pricing for Wonder Full, which is limited to Muse Code. Its cost cannot be directly compared with the API model’s evaluated task cost using the information currently available.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
