🔍 Read the full analysis: Muse Wonder Full Heads To The AI Frontier, With Muse Wonder Charging on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Artificial Analysis’ October 12 evaluation gives Meta’s Muse Wonder Full a score of 56, while the API-available Muse Wonder scores 52. The reported cost for one Intelligence Index task is $4.52 for 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 per-token price disclosed.
Meta’s new Muse Wonder Full scored 56 on Artificial Analysis’ Intelligence Index, tying Claude Sonnet 5.5 and placing behind Claude Opus 5.5 among the 27 configurations compared in the source report. The API-available Muse Wonder scored 52, but Artificial Analysis estimated a task cost of $4.52—far above GPT-6.1 Sol’s $0.72 at the same index score.
Artificial Analysis published its evaluation on October 12, 2026, six weeks after the release of Muse Spark 1.3, according to the source article by Thorsten Meyer. Muse Wonder Full’s score of 56 was eight points above Spark 1.3’s 48, and the evaluator reported improvement across all ten assessments in the index. Among the 27 model configurations compared, Claude Opus 5.5 in its max configuration scored higher, at 58; Claude Sonnet 5.5 in its max configuration also scored 56.
The two new models are proprietary, accept text, images, speech or audio, and video, and have a reported one-million-token context window. Both return text. Their access differs: Muse Wonder is offered through Meta’s API and in Muse Code, while Muse Wonder Full is available only to Muse Code users. Meta’s listed Muse Wonder rates are $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens.
The source report says Muse Wonder Full placed in the top three on six of the ten index evaluations. It reports a 57% result on Terminal-Bench 4.0, up from 33% for Spark 1.3, and 1,857 Elo on GDPval-AA, an assessment of economically useful knowledge work. Muse Wonder, the version available through the API, recorded 52% on Terminal-Bench and 68% on AutomationBench-AA, up from 58% for Spark 1.3. These are evaluator results, not guarantees of performance in a particular workplace.
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
The Cost Behind Muse Wonder’s Score
The release presents a gap between model capability and API economics. Artificial Analysis assigned Muse Wonder a score of 52, the same as GPT-6.1 Sol, but estimated one index task at $4.52 for Meta’s model and $0.72 for Sol. The source also reports costs of $1.99 for Gemini 4 Argon, which scored 53, and $3.26 for GPT-6 Astra, also at 53. These figures apply to the evaluator’s index tasks; they are not universal prices for every prompt or workload.
The reported token use helps explain the cost difference. Muse Wonder consumed 7.69 million input tokens per task, compared with 6.03 million for Spark 1.3, according to the article. Artificial Analysis attributed $3.43, or 76% of the $4.52 task cost, to input. The article says Muse Wonder Full used 6.44 million input tokens, but it has no disclosed per-token price because it is not sold through the API. Buyers therefore cannot compare its task economics on the same published basis.
For developers choosing a model, benchmark scores are only one part of the decision. A model’s cost can depend on how many tokens it processes, the prices charged for those tokens and the reliability required for the job. The source also reports hallucination rates of 27% for Muse Wonder and 23% for Muse Wonder Full, compared with 33% for Spark 1.3. Those results indicate progress in the evaluator’s measure, while leaving verification relevant for tasks where incorrect answers carry costs.
As an affiliate, we earn on qualifying purchases.
Two Models, Different Access
The source article describes Muse Wonder and Muse Wonder Full as Meta’s follow-up to Muse Spark 1.3, released six weeks earlier. Spark scored 48 on the Intelligence Index. The new results indicate a substantial improvement in the evaluator’s overall measure, but the stronger model is restricted to Muse Code rather than being available as a priced API model.
Artificial Analysis also compared coding-agent setups, which combine a model with the software harness it runs in. The source reports Muse Code with Muse Wonder Full at 68 on its Coding Agents Index, tied with Claude Code using Claude Sonnet 5.5. Claude Code with Opus 5.5 scored 66, while Muse Code with the API model Muse Wonder scored 63. Because this index tests model-and-harness combinations, the result does not establish how Wonder Full would perform inside a different coding environment.
The article’s comparisons reflect the configurations and evaluation results it lists, rather than a claim that one model leads every task. For example, its reported AA-Omniscience scores were 30 for Wonder Full and 25 for Wonder, compared with 46 for Claude Opus 5.5. Artificial Analysis’ index aggregates different kinds of tests; individual results can matter more or less depending on how a model will be used.
What the Benchmarks Leave Open
The source does not provide a per-token price for Muse Wonder Full, so its cost per task cannot be compared directly with API models from the information given. Nor does the report specify how Meta’s prices or the evaluator’s token consumption would translate into the cost of a customer’s own workload.
Benchmark results also do not establish performance across every use case. The article reports a 23% hallucination rate for Wonder Full and 27% for Wonder, but does not give details here about the rate’s uncertainty or how results vary by task. It remains unclear whether Meta plans to make Wonder Full available through its API, publish a price for it, or change Wonder’s listed rates. The source material does not include a response from Meta on those points.
Pricing and Wider Availability
The next developments to watch are whether Meta expands access to Muse Wonder Full and whether it publishes a price that allows customers to assess its operating costs. For the API model, potential customers can compare Meta’s listed token rates with the task-cost estimates and benchmark scores in Artificial Analysis’ evaluation, while testing performance on their own workloads.
For now, the confirmed picture in the source report is two different offerings: a higher-scoring model available only inside Muse Code, and a lower-scoring model available through the API at published rates. Meta has not, in the supplied material, announced further availability or pricing changes.
Key Questions
What did Muse Wonder Full score?
Artificial Analysis rated Muse Wonder Full at 56 on its Intelligence Index v4.3, as reported in the source article. That tied the listed score for Claude Sonnet 5.5 in its max configuration; Claude Opus 5.5 in max scored 58.
How much does Muse Wonder cost?
Meta’s listed API prices, as reported in the article, are $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 is an evaluator-specific cost estimate, not a fixed charge for every request.
Can customers buy Muse Wonder Full through the API?
Not according to the source article. It says Muse Wonder Full is available only to Muse Code users and has no per-token price, while Muse Wonder is available through Meta’s API and in Muse Code.
Does the evaluation show Muse Wonder is the cheapest model at its score?
No. The source reports that Muse Wonder and GPT-6.1 Sol both scored 52, while their estimated Intelligence Index task costs were $4.52 and $0.72, respectively. The figures describe the evaluator’s tasks and do not determine costs for every customer workload.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
