🔍 Read the full analysis: Muse Wonder Full And The AI Frontier: A New Arrival, A Charge 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, two proprietary models that Artificial Analysis evaluated on October 12, 2026. Wonder Full scored 56 on its Intelligence Index, while the API-accessible Wonder scored 52 and cost $4.52 per index task—far more than some similarly scoring competitors.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that scored above their predecessor in an independent evaluation published by Artificial Analysis on October 12. Wonder Full scored 56 on the Intelligence Index v4.3, while the API model Wonder scored 52; Artificial Analysis estimated a Wonder task at $4.52, substantially more than some rivals with similar index scores.
Artificial Analysis reported that Muse Wonder Full improved on all 10 evaluations in its index compared with Muse Spark 1.3, rising from 48 to 56. Among 27 configurations compared in the source report, only Claude Opus 5.5 at 58 scored higher; Claude Sonnet 5.5 also scored 56. Wonder Full placed in the top three on six of the 10 evaluations.
The strongest reported gains included Terminal-Bench 4.0, where Wonder Full scored 57%, up from 33% for Spark 1.3, and GDPval-AA, an evaluation of 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 the results fall within one another’s confidence intervals. In Artificial Analysis’s Coding Agents Index, Muse Code paired with Wonder Full scored 68, tying the listed Claude Code and Sonnet 5.5 configuration.
Muse Wonder, the model offered through Meta’s API and Muse Code, scored 52 on the Intelligence Index. Its list price is $4 per million input tokens and $20 per million output tokens, matching the reported Opus 5.5 list price. Wonder Full is available only to Muse Code users and has no per-token price in the source material, so its cost per task cannot be calculated from public pricing there.
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 API Costs
The release puts Meta’s performance and pricing in tension. Artificial Analysis rated Wonder at the same index score as GPT-6.1 Sol, 52, but estimated that one index task cost $4.52 on Wonder and $0.72 on Sol. The report also put Gemini 4 Argon at 53 for $1.99 per task and GPT-6 Astra at 53 for $3.26. These figures make cost a material part of the comparison for customers weighing models for repeated or high-volume use.
The estimate is not simply a comparison of per-token list prices: the source says Wonder used 7.69 million input tokens per index task, compared with 6.03 million for Spark 1.3, and that input accounted for $3.43, or 76%, of Wonder’s $4.52 task estimate. Its output use was about 54,000 tokens per task. Actual bills can vary with workload and usage, but the reported evaluation suggests that token consumption as well as rates affects the cost comparison.
Performance improvements also do not remove the need to check model outputs. Artificial Analysis reported a 23% hallucination rate for Wonder Full and 27% for Wonder, down from 33% for Spark 1.3, but still above the 15% reported for Gemini 4 Argon among models scoring at least 45 on the index. Those evaluation results are a reason for buyers to test models against their own tasks and retain verification for consequential work; they are not a guarantee of performance in any particular deployment.
As an affiliate, we earn on qualifying purchases.
Two Models, Different Access
The release follows Muse Spark 1.3 by six weeks. Both new models are described in the source as proprietary, with a one-million-token context window and text, image, audio or speech, and video input; both return text. Their availability differs: Wonder can be accessed through Meta’s API and inside Muse Code, while Wonder Full is limited to Muse Code.
The distinction matters when interpreting the evaluation. Artificial Analysis’s Coding Agents Index scores a model together with its software harness. The reported tie between Muse Code plus Wonder Full and Claude Code plus Sonnet 5.5 therefore compares complete configurations; it does not show how Wonder Full would perform inside Claude Code. The source also gives no per-token price for Wonder Full, making its economics difficult to compare with API models.
On knowledge and factual reliability measures, the source reports progress but not leading scores. Wonder Full received 30 on AA-Omniscience and Wonder 25, while the listed scores for Opus 5.5, GPT-6 Astra and Gemini 4 Argon were 46, 43 and 42. Artificial Analysis’s index and task-cost figures are evaluation results reported by the source, not independent evidence of how a model will perform for every user.
As an affiliate, we earn on qualifying purchases.
Costs and Real-World Results
Wonder Full’s per-task cost is not available in the supplied material because Meta offers it only inside Muse Code and no per-token price is listed. The $4.52 figure for Wonder is an Artificial Analysis estimate for an index task, not a fixed price for every customer request; the source does not specify how closely that benchmark workload matches typical production use.
The evaluation figures also do not establish how models will perform across different prompts, tools, or business settings. The supplied report does not include a statement from Meta explaining its pricing, the expected availability of Wonder Full through an API, or whether the company plans changes to access or rates. Those points remain unconfirmed here.
AI performance evaluation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Buyers Should Watch
The next practical test is whether Meta expands access to Wonder Full or publishes pricing that allows customers to compare its performance with API models. For now, buyers can access Wonder through Meta’s API and Muse Code, while Wonder Full is restricted to Muse Code, according to the source report.
Organizations considering either model will need to test representative tasks, measure their own token use and costs, and check accuracy against their requirements. The reported benchmarks offer a comparison point, but the models’ performance and total cost in a specific workflow remain to be established.
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, image, audio or speech, and video input and return text, according to the source report.
How did Muse Wonder Full score?
Artificial Analysis gave Wonder Full 56 on Intelligence Index v4.3, compared with 48 for Muse Spark 1.3. The source says it improved on all 10 evaluations in the index.
How much does Muse Wonder cost?
Meta’s listed API rates in the source are $4 per million input tokens and $20 per million output tokens. Artificial Analysis estimated $4.52 for one index task; that is an evaluation-specific estimate, not a universal charge per request.
Can customers buy Muse Wonder Full through the API?
The source report says Wonder Full is available only to Muse Code users and has no listed per-token price. It does not confirm whether Meta plans broader access.
Does a lower hallucination rate mean the models are reliable without checks?
No. The source reports lower hallucination rates than Muse Spark 1.3, but gives rates of 23% for Wonder Full and 27% for Wonder in the cited comparison. Those evaluation results do not guarantee accuracy on a particular task.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
