🔍 Read the full analysis: Muse Wonder Full Joins The AI Frontier—What Comes Next? on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Meta has released Muse Wonder and Muse Wonder Full, two proprietary models with a 1-million-token context window. Artificial Analysis scored Wonder Full at 56 and the API-available Wonder at 52, while its benchmark cost estimate for Wonder was $4.52 per task—far above some models with similar scores.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that scored 52 and 56, respectively, on Artificial Analysis’s Intelligence Index v4.3. The evaluation, published October 12, puts Wonder Full near the top of the benchmark, while estimating that one index task using the API-available Wonder costs $4.52—more than comparable-scoring alternatives cited in the report.
Artificial Analysis reports that Muse Wonder Full scored 56, eight points above Muse Spark 1.3’s 48, and improved on all ten evaluations in the index. Among the 27 configurations compared in the source, only Claude Opus 5.5 at its maximum setting scored higher, at 58; Claude Sonnet 5.5 at its maximum setting also scored 56. Wonder Full placed in the top three on six of the ten evaluations.
The strongest reported change was on Terminal-Bench 4.0, a test of completing work in a shell: Wonder Full scored 57%, compared with 33% for Muse Spark 1.3. It also scored 1,857 Elo on GDPval-AA, an evaluation of economically useful knowledge work. Artificial Analysis places that result between Claude Opus 5.5 at 1,866 and Claude Sonnet 5.5 at 1,838, with the results described as within confidence intervals.
Muse Wonder scored 52 on the index and is the version available through Meta’s API and in Muse Code, the company’s coding agent. The benchmark’s estimated cost for one index task was $4.52 for Wonder, compared with $0.72 for GPT-6.1 Sol at the same score and $1.99 for Gemini 4 Argon at 53. These are Artificial Analysis estimates for the benchmark task, not quoted prices for a typical user’s workload.
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 High API Costs
The release gives developers and businesses new performance data to weigh against access and cost. Wonder Full’s reported results place it among leading configurations on several evaluations, but it is available only to Muse Code users and has no public per-token price in the source. Users therefore cannot compare its cost per task directly from published token rates.
Wonder is the model customers can access through Meta’s API, but the source’s cost estimate complicates its score. Artificial Analysis calculates that Wonder costs 6.3 times as much as GPT-6.1 Sol for an index task at the same score, and about 2.3 times as much as Gemini 4 Argon, which scores one point higher. Those comparisons may help buyers judge value, but they apply to this benchmark’s task mix and should not be treated as universal operating costs.
Reliability is another consideration. Artificial Analysis reports a 27% hallucination rate for Wonder and 23% for Wonder Full, down from 33% for Spark 1.3. The report also says Gemini 4 Argon’s rate is 15%. The results indicate improvement from Meta’s prior model, while leaving a reason for users to verify outputs in workflows where errors carry costs.
As an affiliate, we earn on qualifying purchases.
Two Models, Different Access
Meta released the models about six weeks after Muse Spark 1.3. Both new models are proprietary, have a reported one-million-token context window, and accept text, images, speech or audio, and video as input; both return text, according to the source.
The two versions differ in availability. Muse Wonder is available through Meta’s API and inside Muse Code. The source lists its rates as $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens—the same listed rates as Claude Opus 5.5. Muse Wonder Full is available only to Muse Code users, with no per-token price provided.
Artificial Analysis also compared coding-agent combinations, which include both a model and its software harness. It scored Muse Code with Wonder Full at 68, matching Claude Code with Sonnet 5.5 at 68. The source cautions that this does not establish how Wonder Full would perform inside Claude Code; it measures each model paired with its own coding environment.
“The numbers are the most interesting Meta has posted this year.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
Costs and Reliability Still Need Testing
The source does not provide a per-token price for Muse Wonder Full, so its cost for comparable tasks cannot be calculated from the listed rates. The $4.52 estimate for Wonder reflects Artificial Analysis’s Intelligence Index task, including that benchmark’s token use; costs may differ across applications, prompts, and settings.
The evaluation reports benchmark results, not independent evidence of performance across every real-world use. The source gives no detailed information about how Meta’s API pricing or model availability may change, nor does it establish how often the models will make errors in a particular customer’s workflow. The reported hallucination rates are benchmark measurements and do not predict the error rate for every use case.
Some comparisons also involve configurations with different settings or model-and-harness combinations. The source says Wonder Full’s GDPval-AA result is within the confidence intervals of the cited Claude scores, so the results do not establish a definite ranking among those models on that evaluation.
As an affiliate, we earn on qualifying purchases.
API Use Will Test the Value
The next practical test is whether developers adopt Muse Wonder despite the reported benchmark-task cost, and how its performance and spending compare on their own workloads. API users can evaluate the listed token rates against actual usage, including how much input the model processes, rather than relying only on a single index estimate.
For Wonder Full, wider cost comparisons will depend on whether Meta provides additional pricing or access beyond Muse Code. Further independent evaluations could also clarify how consistently the models perform across coding, knowledge work, and other tasks, and how their reported hallucination rates translate to specific applications. The source does not give a schedule for any such changes or evaluations.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Meta release?
Meta released Muse Wonder and Muse Wonder Full, proprietary models with a reported one-million-token context window. Both accept text, images, audio or speech, and video as input, and return text, according to the source.
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 available only to Muse Code users, according to the source.
How did the models score?
Artificial Analysis scored Wonder Full at 56 and Wonder at 52 on Intelligence Index v4.3. The scores are benchmark results and do not establish performance on every task.
What does the $4.52 cost estimate mean?
It is Artificial Analysis’s estimated cost for one Intelligence Index task using Muse Wonder. It is not a guaranteed price for every API request or a measure of what all users will pay for their workloads.
What remains unknown about Wonder Full?
The source gives no per-token price for Wonder Full, which is available only in Muse Code. Its cost on comparable tasks and whether Meta will broaden access are not specified.
Source: ThorstenMeyerAI.com
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
