🔍 Read the full analysis: Muse Wonder Full: A Fresh Arrival On The AI Frontier on ThorstenMeyerAI.com
Get audio and creator gear 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 with a one-million-token context window. Artificial Analysis scored the Full version at 56 and the API model at 52, but estimated one Intelligence Index task costs $4.52 on Muse Wonder, compared with $0.72 for GPT-6.1 Sol at the same score. Muse Wonder Full is available only through Muse Code, with no published per-token price.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that improve on its Muse Spark 1.3 system in independent benchmark testing, but the version sold through Meta’s API carries a high estimated task cost. Artificial Analysis scored Muse Wonder Full at 56 on its Intelligence Index v4.3 and Muse Wonder at 52; its estimate for one index task on the API model was $4.52, compared with $0.72 for GPT-6.1 Sol, which also scored 52.
Artificial Analysis published its evaluation on October 12, six weeks after the release of Muse Spark 1.3, according to the source report. Muse Wonder Full’s score of 56 is eight points above Spark 1.3’s 48. The evaluator said Wonder Full improved on all ten assessments in the index and placed among the top three on six of them. Among 27 model configurations compared in the report, Claude Opus 5.5 in its max configuration scored higher, at 58; Claude Sonnet 5.5 max also scored 56.
The versions differ in access. Muse Wonder is available through Meta’s API and in the Muse Code coding agent, at listed prices of $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Muse Wonder Full is limited to Muse Code and has no per-token price, so its cost per task cannot be calculated from the published information. Both models have a one-million-token context window, accept text, images, audio and video as input, and return text.
Artificial Analysis reported several gains for Wonder Full, including a 57% Terminal-Bench 4.0 score, up from 33% for Spark 1.3, and a 71% score on AutomationBench-AA. On GDPval-AA, which evaluates economically useful knowledge work, Wonder Full scored 1,857 Elo, between Claude Opus 5.5 at 1,866 and Sonnet 5.5 at 1,838; the report said the results were within the models’ confidence intervals. Muse Wonder, the API version, scored 52% on Terminal-Bench and 68% on AutomationBench-AA.
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 puts Meta in a stronger position in evaluations of coding and knowledge-work tasks. Artificial Analysis also scored the combination of Muse Code and Muse Wonder Full at 68 in its Coding Agents Index, matching Claude Code with Sonnet 5.5. That index assesses the model and its software harness together; it does not establish that Wonder Full would perform the same way inside Claude Code or that the underlying models are interchangeable.
For customers choosing an API model, the reported benchmark result comes with a cost question. Artificial Analysis estimated $4.52 per Intelligence Index task for Muse Wonder, compared with $0.72 for GPT-6.1 Sol at the same score and $1.99 for Gemini 4 Argon, which scored 53. These are evaluator estimates for the benchmark task, not a fixed charge for every customer workload. Actual bills can vary with usage, but the comparison highlights a potential obstacle for developers weighing performance against operating cost.
Reliability also remains relevant. Artificial Analysis put the hallucination rate at 23% for Wonder Full and 27% for Wonder, down from 33% for Spark 1.3. In the report’s comparison of models scoring at least 45 on the Intelligence Index, Gemini 4 Argon’s reported rate was 15%. The figures do not mean every answer has those odds of being wrong, but they support retaining checks for work where an incorrect response could have financial or operational consequences.
As an affiliate, we earn on qualifying purchases.
How the Two Models Differ
The product split shapes what the benchmark scores mean in practice. Wonder Full is the higher-scoring system, but users cannot access it as a separately priced API model based on the information in the report. It is offered through Muse Code, Meta’s coding agent. Wonder is the model developers can call through Meta’s API, and it scored four points lower on the Intelligence Index.
Artificial Analysis’ task-cost estimate reflects the tokens used in its benchmark, not just Meta’s listed token rates. The report says Muse Wonder used 7.69 million input tokens per index task, compared with 6.03 million for Muse Spark 1.3; it produced about 54,000 output tokens. The evaluator attributed $3.43, or 76% of the estimated $4.52 task cost, to input. This helps explain why the benchmark’s estimated cost is high, although it does not by itself establish how much a typical customer’s task would cost.
The report also compares token prices: Muse Spark 1.3 was listed at $1.25 per million input tokens and $4.25 per million output tokens, while Wonder is listed at $4 and $20. Those rates match the reported prices for Claude Opus 5.5. The comparisons are based on the source’s stated prices and evaluation methods; they are not a guarantee of equivalent performance or cost on every workload.
“The difference is the bill: one Intelligence Index task costs $4.52 on Muse Wonder and $0.72 on Sol.”
— Thorsten Meyer, ThorstenMeyerAI.com
Pricing and Reliability Questions
Meta has not published a per-token price for Muse Wonder Full, which is available only through Muse Code. The source material does not state whether Meta plans to offer it through an API, how access may change, or what it costs Meta to serve. Without those details, buyers cannot compare its per-task price directly with API models.
The $4.52 figure is an Artificial Analysis estimate for one benchmark task, based on the tokens used in that evaluation. It is not a universal price for a customer request, and the report’s figures do not establish costs across different prompts, task lengths or production workloads. The source also does not provide enough information to independently assess how representative the index tasks are of each reader’s use case.
The reported hallucination rates and benchmark scores offer comparative evidence, not guarantees of performance in real deployments. It remains unclear how the models will perform across different languages, industries and longer-running tasks, and whether the gains will hold as users move beyond the evaluator’s test set.
API Access and Real-World Costs
Developers can test Muse Wonder through Meta’s API or use it in Muse Code at the published token rates. Muse Wonder Full’s next practical test is whether its strong evaluation results translate into dependable performance for coding and other work inside Muse Code. The report does not specify a date for wider access or a separate pricing announcement.
For organizations considering the API model, the next step is to measure their own workloads: token use, answer quality, error rates and total cost. That would help determine whether Muse Wonder’s benchmark performance justifies its estimated expense for a particular application. Readers should also watch for further model evaluations and any Meta update on Wonder Full access or pricing.
Key Questions
What did Meta release?
Meta released Muse Wonder and Muse Wonder Full, proprietary models with a one-million-token context window. Both accept text, images, audio and video as input and return text.
Which model can developers access through an API?
Muse Wonder is available through Meta’s API and in Muse Code. Muse Wonder Full is available only through Muse Code, according to the source report.
How much does Muse Wonder cost?
Meta’s listed price is $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Artificial Analysis estimated $4.52 for one of its Intelligence Index tasks; that estimate is not a fixed price for every customer task.
How did the models score?
Artificial Analysis gave Muse Wonder Full a score of 56 and Muse Wonder a score of 52 on its Intelligence Index v4.3. Muse Spark 1.3 scored 48 in the comparison.
Is Muse Wonder Full available at a published API price?
No per-token price for Muse Wonder Full is provided in the source material. It says the model is available only to Muse Code users, leaving its per-task cost unclear.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
