🔍 Read the full analysis: Muse Wonder Full Enters AI’s Frontier As Muse Wonder Gets Powered Up 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
Meta has released Muse Wonder and Muse Wonder Full, two proprietary models with one-million-token context windows. Artificial Analysis rated Wonder Full 56 on its Intelligence Index, while API-available Wonder scored 52; its estimated cost per index task was $4.52, compared with $0.72 for GPT-6.1 Sol at the same score. The evaluation is independent testing, but the figures do not establish how the models perform across every real-world use.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that accept text, images, audio and video and have one-million-token context windows. In an evaluation published Oct. 12 by Artificial Analysis, Wonder Full scored 56 on the firm’s Intelligence Index, while Wonder, the version offered through Meta’s API, scored 52. The testing puts the Full model near leading competitors, but the API model’s estimated cost per benchmark task was $4.52—far above GPT-6.1 Sol’s $0.72 for the same index score.
Artificial Analysis rated Muse Wonder Full at 56 on its Intelligence Index v4.3, eight points above Meta’s Muse Spark 1.3, which scored 48. The evaluator said Wonder Full improved on all ten assessments in the index and ranked in the top three on six. Among the 27 configurations compared in the source report, Claude Opus 5.5 (max) scored higher at 58; Claude Sonnet 5.5 (max) also scored 56.
The two releases are not equally available. Muse Wonder is accessible through Meta’s API and Muse Code, the company’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 in Muse Code and has no per-token price listed in the source material, preventing an equivalent public cost comparison.
Artificial Analysis also reported a 68 score for the combination of Muse Code and Wonder Full in its Coding Agents Index, equal to Claude Code paired with Claude Sonnet 5.5. The index assesses model-and-harness combinations, so it does not show how Wonder Full would perform inside Claude Code. On the API model, the source gives a 52% Terminal-Bench result, 68% on AutomationBench-AA and a 27% hallucination rate; those figures are benchmark results, not guarantees of performance in customer workflows.
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 of Meta’s API Model
The release gives Meta a credible showing in independent tests, while leaving a practical question for organizations choosing a model: how much performance they get for the price. Artificial Analysis estimated that one Intelligence Index task cost $4.52 with Muse Wonder. GPT-6.1 Sol scored the same 52 for $0.72, while Gemini 4 Argon scored 53 for $1.99. These are costs for the evaluator’s benchmark workload, not fixed prices for every user task.
The comparison matters because API access allows developers to test models in their own software and pay according to usage. Meta’s published Wonder rates—$4 per million input tokens and $20 per million output tokens—match the Opus 5.5 list prices cited in the source report. The benchmark comparison suggests that price and score do not line up neatly for this particular evaluation: Wonder’s score matches Sol’s, while its estimated task cost is higher. Buyers may weigh other factors, such as integration, output quality on their own work and data requirements, which this index does not settle.
Wonder Full’s stronger result matters for Meta’s coding environment, but its limited availability and unpublished per-token cost make it harder for outside buyers to compare directly. The Coding Agents Index result supports the claim that the Muse Code-plus-model pairing performed well on that test; it does not establish that the model alone is equivalent to other systems in different coding tools.
As an affiliate, we earn on qualifying purchases.
From Muse Spark to Wonder
The release follows Muse Spark 1.3 by about six weeks, according to the source report. Artificial Analysis’ comparison gives the new models higher results on several listed measures. Wonder Full’s Terminal-Bench score was 57%, against 33% for Spark 1.3. Wonder scored 52% on the same assessment. Terminal-Bench evaluates tasks performed in a shell environment; a benchmark gain is evidence of improvement on that test, not proof of a general change in agent capability.
The source also reports that Wonder Full scored 1,857 Elo on GDPval-AA, an assessment of economically useful knowledge work. That falls between Claude Opus 5.5 at 1,866 and Claude Sonnet 5.5 at 1,838; the report says the results are within each other’s confidence intervals. On AA-Omniscience, which rewards correct answers and penalizes confident errors, Wonder Full scored 30 and Wonder scored 25. The report lists higher results for Claude Opus 5.5, GPT-6 Astra and Gemini 4 Argon.
Artificial Analysis estimated Wonder’s benchmark-task cost at $4.52, compared with $1.60 for Spark 1.3. Its token accounting attributes $3.43, or 76%, to input, and says Wonder processed 7.69 million input tokens per task. Wonder Full processed 6.44 million. These are evaluator-specific estimates; actual bills depend on usage and the applicable pricing and token counts.
“The difference is the bill.”
— Thorsten Meyer, ThorstenMeyerAI.com
Pricing and Real-World Performance
The available information does not give a public per-token price for Wonder Full, so its cost cannot be compared with API models on the same basis. Nor does a single benchmark-task estimate show what a developer’s bill would be for a different workload. Token use, task design and model settings can change costs.
The source reports Artificial Analysis results but does not provide the full evaluation methodology, confidence intervals for every measure or independent results from other testing organizations. Its coding index scores combine a model with a software harness, meaning they cannot isolate the model’s contribution. It is also unclear from the supplied material when broader access to Wonder Full might be offered, whether its pricing or availability will change, or how either model performs on specific customer applications.
Hallucination figures warrant care as well. Artificial Analysis reportedly measured 23% for Wonder Full and 27% for Wonder, compared with 33% for Spark 1.3; the source also gives 15% for Gemini 4 Argon. Those rates depend on the evaluator’s test and should not be treated as the odds of an error in every use. The report itself advises retaining verification for workflows where a confident wrong answer could cause harm.
What API Buyers Can Test
Developers can access Muse Wonder through Meta’s API and test it in Muse Code, while Wonder Full is presently limited to Muse Code users, according to the source material. For API buyers, the next practical step is to compare the model on their own prompts and workloads, tracking both answer quality and token consumption rather than relying only on a benchmark score.
Further comparisons will depend on additional independent evaluations, clearer information about Wonder Full’s cost and availability, and evidence from use beyond the tests cited here. Meta’s next release or pricing update is not specified in the source material, so no timing for such a development is confirmed.
Key Questions
What did Meta release?
Meta released Muse Wonder and Muse Wonder Full, proprietary models that accept text, images, audio and video and have one-million-token context windows. Wonder is available through Meta’s API and Muse Code; Wonder Full is available only through Muse Code, according to the source report.
How did the models score?
Artificial Analysis rated Wonder Full 56 and Wonder 52 on its Intelligence Index v4.3. The same report lists Muse Spark 1.3 at 48. These scores describe results on that index and do not predict performance on every task.
How much does Muse Wonder cost?
Meta’s listed price, as given in the source, 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 reflects its benchmark workload and is not a universal task price.
Can customers buy Muse Wonder Full through the API?
The supplied report says Wonder Full is available only to Muse Code users and has no per-token price. It does not establish whether Meta plans to add API access or when that might happen.
Does the benchmark establish that Muse Wonder is better value?
No. Artificial Analysis reported that Wonder scored 52 at an estimated $4.52 per index task, while GPT-6.1 Sol scored 52 at $0.72. That comparison is relevant to the evaluator’s workload, but buyers need to test their own tasks and account for their usage before judging value.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
