🔍 Read the full analysis: Muse Wonder Full Joins AI’s Frontier: What’s The Muse Wonder Connection? on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- 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, six weeks after Muse Spark 1.3. Artificial Analysis rates Wonder Full at 56 on its Intelligence Index, but it is available only through Muse Code; the API model scores 52 and costs $4.52 per benchmark task in the evaluator’s comparison.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that mark a substantial benchmark gain over Muse Spark 1.3, according to an evaluation published October 12 by Artificial Analysis. The stronger model scored 56 on the evaluator’s Intelligence Index, while the version available through Meta’s API scored 52; Wonder Full, however, is limited to Meta’s Muse Code product.
Artificial Analysis said Muse Wonder Full improved on all 10 evaluations in its Intelligence Index v4.3 compared with Muse Spark 1.3, rising from 48 to 56. Among 27 configurations included in the source’s comparison, only Claude Opus 5.5 at its maximum setting scored higher, at 58; Claude Sonnet 5.5 at its maximum setting also scored 56. These are evaluator results, not a guarantee of performance in every use case.
The gains included a rise on Terminal-Bench 4.0 from 33% for Spark 1.3 to 57% for Wonder Full. The model scored 1,857 Elo on GDPval-AA, a knowledge-work evaluation, compared with 1,866 for Opus 5.5 and 1,838 for Sonnet 5.5; the source says those results fall within the models’ confidence intervals. In Artificial Analysis’s Coding Agents Index, Muse Code paired with Wonder Full scored 68, matching the listed Claude Code and Sonnet 5.5 configuration.
Muse Wonder is the API-accessible model and is also available inside Muse Code. Meta’s listed price is $4 per million input tokens and $20 per million output tokens, with cached input priced at $0.20 per million tokens. Artificial Analysis’s comparison estimates $4.52 to run one Intelligence Index task with Wonder. That is the same score as GPT-6.1 Sol, which the comparison estimates at $0.72 per task. The figures reflect the evaluator’s benchmark workload and should not be read as a universal cost for all uses.
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
Benchmark Strength Meets API Cost
The release presents two different questions for customers: how capable the models are and what access costs. Wonder Full’s score and coding-agent result put Meta’s own coding product alongside leading configurations in this particular evaluation. But because the model is available only through Muse Code and has no per-token price in the source material, outside users cannot compare its cost per task directly.
For developers choosing an API, the relevant product is Muse Wonder, not Wonder Full. Artificial Analysis scores it at 52, level with GPT-6.1 Sol in the reported index comparison, but estimates a much higher benchmark-task cost. That does not establish that Wonder is poor value across all applications: token use, task type, latency, and output quality can change the economics. It does mean buyers have reason to test their own workloads rather than infer value from the headline score alone.
Reliability also remains relevant. The source reports a 27% hallucination rate for Wonder and 23% for Wonder Full, compared with Spark 1.3’s 33%. The reduction is progress in the cited measure, but it does not remove the risk of confident errors. For workflows where mistakes have financial or operational consequences, independent checks remain a practical safeguard.
As an affiliate, we earn on qualifying purchases.
From Spark to Two Muse Models
The release follows Muse Spark 1.3 by six weeks, according to the source article. Meta has split the new offering into a model sold through an API and a stronger version restricted to Muse Code. Both are described as proprietary, with a one-million-token context window and text, image, speech or audio, and video inputs; both return text.
The source compares Wonder’s API prices with those of other models, noting that its $4 input and $20 output rates match the listed prices for Claude Opus 5.5. It gives $2 per million input tokens and $10 per million output tokens for GPT-6.1 Sol, Gemini 4 Argon, and Claude Sonnet 5.5. Artificial Analysis’s task-cost estimates also reflect how many tokens each model used in its benchmark: the source says Wonder consumed 7.69 million input tokens per task, while Wonder Full consumed 6.44 million. Those workload-specific differences help explain why list prices alone do not determine a benchmark’s total cost.
“The stronger model is the one you can’t buy by the token, and the model you can buy competes one tier lower.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
Access, Costs and Real-World Results
The source does not provide a per-token price or a comparable per-task cost for Muse Wonder Full. It is therefore not possible from these figures to judge its cost efficiency against API models or to know what broader access might cost. The material also does not establish whether Meta plans to make Wonder Full available beyond Muse Code.
Artificial Analysis’s scores and task-cost estimates describe a defined evaluation, not every customer’s workload. The source does not provide enough detail here to independently assess the benchmark’s full methodology, uncertainty ranges for each result, or how performance varies across industries and languages. The reported hallucination rates likewise depend on the evaluator’s test and should not be treated as a universal probability of error.
Meta’s own product statements and pricing are summarized in the source article, but no direct Meta quotation is supplied. Claims about the models’ comparative standing and token consumption should accordingly be understood as findings or calculations attributed to Artificial Analysis through that report, rather than as guarantees from the company.
AI performance evaluation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Buyers Should Watch
The next useful evidence will come from real-world testing and clearer access terms: whether Meta expands Wonder Full availability, publishes a way to compare its costs, or updates the API model’s pricing and capabilities. Developers considering Muse Wonder can run representative tasks against alternatives and track both token use and error rates, since the reported benchmark cost may not match their own usage.
Further independent evaluations may also show whether the reported gains hold across additional tasks and model versions. Until then, the confirmed picture is narrower: Artificial Analysis recorded a strong score for the Muse Code-only model and a score of 52 for API-accessible Wonder, while its benchmark estimates put Wonder’s task cost above several similarly scored competitors.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are Muse Wonder and Muse Wonder Full?
They are two proprietary models released by Meta. Both accept text, images, audio or speech, and video as input and return text, according to the source. Wonder is available through Meta’s API and in Muse Code; Wonder Full is available only to Muse Code users.
How did Muse Wonder Full score?
Artificial Analysis gave it a 56 on Intelligence Index v4.3, eight points above Muse Spark 1.3’s 48. In the comparison cited by the source, Claude Opus 5.5 scored 58 and Claude Sonnet 5.5 also scored 56 at its maximum setting.
How much does Muse Wonder cost?
Meta’s listed API price is $4 per million input tokens and $20 per million output tokens; cached input is listed at $0.20 per million tokens. Artificial Analysis estimated $4.52 to run one task in its Intelligence Index comparison. That task estimate is tied to the evaluator’s workload, not a fixed price for every request.
Can customers buy Muse Wonder Full through the API?
Not according to the source material. It says Wonder Full is available only through Muse Code and has no per-token price, leaving its comparable API cost unknown.
Do the benchmark results mean the models do not make mistakes?
No. The source reports hallucination rates of 23% for Wonder Full and 27% for Wonder in the cited evaluation. Those figures are test-specific, and the report says they do not remove the need to verify outputs in consequential workflows.
Source: ThorstenMeyerAI.com
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
