🔍 Read the full analysis: Muse Wonder Full’s AI Frontier Arrival: Muse Wonder Charges, Too on ThorstenMeyerAI.com
Get smart everyday buys 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, two proprietary models that Artificial Analysis evaluated on October 12. Wonder Full scored 56 on the Intelligence Index, while the API-available Wonder scored 52; the source reports that one benchmark task with Wonder cost $4.52, compared with $0.72 for GPT-6.1 Sol at the same score.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that scored 52 and 56, respectively, on Artificial Analysis’ Intelligence Index v4.3. The results put Wonder Full near leading models on several evaluations, while the API version’s reported $4.52 cost per index task was substantially higher than GPT-6.1 Sol’s $0.72 for the same score of 52, according to the evaluation summarized by ThorstenMeyerAI.com.
Artificial Analysis published its evaluation on October 12, comparing the new models with other configurations on its Intelligence Index. Muse Wonder Full scored 56, up from 48 for Muse Spark 1.3. The source says Wonder Full improved on all ten index evaluations and placed in the top three on six. Muse Wonder scored 52, matching the reported score for GPT-6.1 Sol at its maximum setting.
The models differ in availability. Muse Wonder is available through Meta’s API and in Muse Code, Meta’s coding agent. The source lists its API pricing as $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Muse Wonder Full is available only to Muse Code users and has no per-token price listed, leaving its cost per task unavailable to outside users.
Among Wonder Full’s reported results, Terminal-Bench 4.0 rose to 57%, from 33% for Muse Spark 1.3. The source also reports 1,857 Elo on GDPval-AA, a measure of economically useful knowledge work, and a 33% score on GDP.pdf, tied for the top result in the comparison described. These are benchmark findings, not evidence by themselves of performance in every real-world deployment.
In Artificial Analysis’ Coding Agents Index, Muse Code with Wonder Full scored 68, tying the listed Claude Code and Sonnet 5.5 configuration. The index assesses the model together with its coding harness, so the result does not establish how Wonder Full would perform inside a different coding product.
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 Meet a Higher API Bill
The release matters because Meta’s evaluation results point to a narrower performance gap with leading commercial models, but access and price are uneven across its two offerings. Wonder Full is the higher-scoring model, yet it is restricted to Muse Code and has no published per-token rate in the supplied material. The model developers can price and compare directly through the API, Wonder, scored lower.
Artificial Analysis’ estimated cost for one Intelligence Index task puts that trade-off in sharper relief: Wonder cost $4.52, against $0.72 for GPT-6.1 Sol, which also scored 52. The estimate is tied to the benchmark’s workload and token use; it is not a universal price for a typical user request. Still, it gives developers a reason to compare both quality and expected usage before choosing a model.
The evaluation also reports a 27% hallucination rate for Wonder and 23% for Wonder Full, down from 33% for Muse Spark 1.3. The improvement is meaningful in the reported comparison, but the figures indicate that incorrect confident answers remain a concern. Teams using the models for work where errors carry costs may still need review and verification processes.
As an affiliate, we earn on qualifying purchases.
Six Weeks After Muse Spark
The new models arrive six weeks after Muse Spark 1.3. In the source’s comparison, Wonder Full’s Intelligence Index score rose from 48 to 56, while Wonder reached 52. The source also reports that Wonder improved on Spark 1.3 in several task measures, including AutomationBench-AA, where Wonder scored 68%, compared with 58% for its predecessor.
Both models are described as accepting text, images, speech or audio, and video, returning text, and supporting a one-million-token context window. The available details do not establish how reliably every input type performs across different tasks. The central distinction is access: users can call Wonder through Meta’s API, while Wonder Full is confined to Muse Code.
Artificial Analysis’ cost comparison reflects the token consumption measured on its Intelligence Index tasks. The source says Wonder used about 7.69 million input tokens per task and about 54,000 output tokens. It attributes $3.43 of the $4.52 estimated task cost to input. Those benchmark-specific totals help explain the estimate, but actual costs can vary with prompts, workload, caching and usage patterns.
“Meta closed the capability gap. It did not close the price gap.”
— 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.
Costs and Real-World Accuracy
The supplied material does not include a per-token price for Muse Wonder Full, so its cost cannot be compared directly with API models on the same basis. It also does not provide independent deployment data showing how the benchmark results translate to customers’ workloads, latency, reliability or total operating costs.
The $4.52 figure is an Artificial Analysis estimate for one Intelligence Index task, not a fixed charge for every interaction. The source gives token counts and a cost breakdown, but the exact bill for a user will depend on the prompts and tasks involved. Benchmark scores and hallucination rates also describe a particular evaluation; they cannot guarantee results in an individual application.
As an affiliate, we earn on qualifying purchases.
Watch API Access and Pricing
The next practical tests will be whether Meta expands access to Muse Wonder Full or publishes pricing that permits direct cost comparisons. Developers considering Wonder can assess its API rates against their own workload, rather than relying solely on the benchmark’s estimated task cost.
Further independent evaluations could clarify how the models perform across different tasks and whether their reported accuracy and token use hold in routine deployments. Until then, the published results establish a strong benchmark showing for Wonder Full and a higher measured task cost for Wonder in this evaluation, while leaving the full economics and real-world performance unsettled.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Meta release?
Meta released Muse Wonder and Muse Wonder Full, proprietary models that accept text, images, audio or speech, and video, and return text. Both are reported to have a one-million-token context window.
How did the models score?
Artificial Analysis’ Intelligence Index v4.3 gave Wonder Full a score of 56 and Wonder a score of 52. The source reports that Wonder Full improved on all ten evaluations in the index.
Which model can developers access through an API?
Muse Wonder is available through Meta’s API and in Muse Code. Wonder Full is listed as available only to Muse Code users, with no per-token price provided in the source material.
What does the $4.52 cost figure represent?
It is the reported estimated cost to run one Artificial Analysis Intelligence Index task with Muse Wonder. It is not a flat charge for every user prompt, and actual costs depend on the workload and token use.
Do the benchmark results mean the models do not hallucinate?
No. The source reports hallucination rates of 23% for Wonder Full and 27% for Wonder in the cited evaluation. Those results show improvement from Muse Spark 1.3’s reported 33% rate, but they do not eliminate the risk of incorrect answers.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
