🔍 Read the full analysis: Muse Wonder Full Makes A Move Into The AI Frontier 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
Artificial Analysis gave Meta’s Muse Wonder Full a score of 56 on its Intelligence Index, up eight points from Muse Spark 1.3. The stronger model is available only through Meta’s Muse Code, while the API model scores 52 and costs $4.52 per benchmark task—far more than some similarly scoring rivals.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that scored higher than its previous Muse Spark 1.3 on an independent evaluation—but only the lower-scoring model is available through Meta’s API. Artificial Analysis rated Wonder Full 56 and API model Wonder 52 on its Intelligence Index v4.3, while estimating that one benchmark task with Wonder costs $4.52.
Artificial Analysis published its evaluation on October 12, comparing the new models with other leading systems. Wonder Full scored eight points above Muse Spark 1.3’s 48 and improved on all ten evaluations in the index, according to the assessment. Among 27 configurations compared in the source analysis, only Claude Opus 5.5 in its max configuration scored higher, at 58; Claude Sonnet 5.5 max matched Wonder Full at 56.
The two models differ in access. Muse Wonder is available through Meta’s API and inside Muse Code, Meta’s coding agent. Its listed price is $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, leaving its cost per task unclear to outside users.
Artificial Analysis reported that Wonder Full scored 57% on Terminal-Bench 4.0, up from 33% for Muse Spark 1.3, and ranked among the top three on six of the ten evaluations. The API model scored 52% on Terminal-Bench and 68% on AutomationBench-AA. The evaluation also reported hallucination rates of 23% for Wonder Full and 27% for Wonder, compared with 33% for Spark 1.3.
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
The release puts Meta closer to the top tier of AI model evaluations, especially for coding and knowledge-work tasks. Artificial Analysis reported that Muse Code paired with Wonder Full scored 68 on its Coding Agents Index, tying its Claude Code and Sonnet 5.5 configuration. That result measures the model and software harness together; it does not establish how Wonder Full would perform inside Claude Code or another system.
For buyers, the access and pricing split may matter as much as the benchmark scores. The model with the strongest reported results cannot be purchased through a public per-token API, while the API model’s estimated task cost is $4.52. Artificial Analysis put GPT-6.1 Sol at the same Intelligence Index score of 52 for $0.72 per task, and Gemini 4 Argon at 53 for $1.99. Those figures suggest Wonder’s price may be difficult to justify for cost-sensitive workloads, though real expenses depend on usage and the tasks being run.
Accuracy remains another practical concern. Wonder Full’s reported 23% hallucination rate is an improvement on its predecessor, but the source comparison lists Gemini 4 Argon at 15%. For work where unsupported answers carry financial or operational risks, these results support keeping human review or other verification in place.
As an affiliate, we earn on qualifying purchases.
From Spark to Wonder
The two releases arrived about six weeks after Muse Spark 1.3. Both accept text, images, speech or audio, and video as input, return text, and have a context window of one million tokens, according to the source material. They are proprietary models, and their separate access arrangements mean the headline benchmark score does not describe a single model that all users can access in the same way.
Artificial Analysis’s Intelligence Index combines results from multiple evaluations. Individual results in the source include Wonder Full at 1,857 Elo on GDPval-AA, behind Claude Opus 5.5 at 1,866, and 33% on GDP.pdf and CritPt, where it ranked first in the reported comparison. It scored 30 on AA-Omniscience, which the source says rewards correct answers while penalizing confident mistakes. Wonder scored 25 on that measure.
The source analysis also reports a substantial difference in token use. Wonder consumed 7.69 million input tokens per index task, compared with 6.03 million for Spark 1.3, while Wonder Full consumed 6.44 million. The analysis attributes $3.43 of Wonder’s $4.52 estimated task cost—76%—to input tokens. These are benchmark-task estimates, not a fixed charge for every use of the API.
“The capability gap is closed. The price gap is not.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
API Costs and Results
Meta’s own detailed evaluation methodology, pricing rationale and explanation of the gap between Wonder and Wonder Full are not included in the source material. The reported figures come from Artificial Analysis’s evaluation as summarized by Meyer; benchmark scores alone do not show how a model will perform on every customer’s workload.
The per-task price for Wonder Full cannot be compared from the supplied information because Meta offers it only inside Muse Code and does not publish a per-token price. The $4.52 figure for Wonder is an estimate for one Intelligence Index task, not a universal cost per prompt. It is also not clear how Meta’s listed API rates or model availability may change after launch.
Hallucination rates and benchmark rankings depend on the tests and methods used. The source does not provide enough detail to establish how those results translate to a particular deployment, or whether the models’ performance and costs will remain stable as access expands.
As an affiliate, we earn on qualifying purchases.
Watch Access and Pricing
The next practical test is how Meta makes the stronger Wonder Full model available, and whether it publishes pricing or broader access beyond Muse Code. Developers considering Wonder can compare the API’s listed token rates with their expected input volume and test the model on representative tasks before relying on the benchmark estimates.
Further evaluations may clarify whether the reported advantage holds across independent tests and real-world deployments. For now, the confirmed development is a pair of released models, an independent score of 56 for Wonder Full and 52 for Wonder, and a notable cost difference between Meta’s API option and some similarly scoring alternatives.
AI development coding environments
As an affiliate, we earn on qualifying purchases.
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 as input. Both have a reported one-million-token context window and return text.
How did the models score?
Artificial Analysis rated Muse Wonder Full 56 and Muse Wonder 52 on Intelligence Index v4.3. The previous model, Muse Spark 1.3, scored 48 in the comparison.
Can developers access Muse Wonder Full through an API?
Not according to the supplied report. Wonder Full is available only inside Muse Code; Muse Wonder is the model available through Meta’s API and Muse Code.
How much does a Muse Wonder benchmark task cost?
Artificial Analysis’s estimate is $4.52 per Intelligence Index task. This is a benchmark-specific estimate, not a fixed price for every API request. Meta’s listed rates are $4 per million input tokens and $20 per million output tokens, with cached input at $0.20 per million tokens.
Is Muse Wonder Full cheaper than competing models?
The supplied material does not give a per-token price or per-task cost for Wonder Full. It reports that API model Muse Wonder cost more per benchmark task than GPT-6.1 Sol and Gemini 4 Argon, despite scoring similarly or slightly lower in that comparison.
Source: ThorstenMeyerAI.com
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
