🔍 Read the full analysis: Muse Wonder Full Joins The Frontier As Muse Wonder Charges Like Before on ThorstenMeyerAI.com
Get the latest gadgets 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
Artificial Analysis rated Meta’s Muse Wonder Full at 56 on its Intelligence Index and the API-accessible Muse Wonder at 52. The evaluation shows strong gains over Muse Spark 1.3, while its cost estimate for an Intelligence Index task puts Wonder well above several models with similar or higher scores.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that received scores of 52 and 56, respectively, in Artificial Analysis’s Intelligence Index v4.3 evaluation published October 12. The stronger model is available only through Meta’s Muse Code coding agent, while the API-accessible Wonder model received a lower score and a reported task cost of $4.52—a price that raises questions about its value against similarly rated rivals.
Artificial Analysis said Muse Wonder Full’s score of 56 was eight points above Muse Spark 1.3’s 48. It placed Wonder Full behind Claude Opus 5.5, which scored 58, and level with Claude Sonnet 5.5. The evaluation also reported gains across all ten index evaluations. These are results from Artificial Analysis’s benchmark, not a guarantee of performance in every real-world task.
The two models share a one-million-token context window and accept text, images, audio or speech, and video as input; both return text. Meta makes Wonder available through its API and in Muse Code. The listed API rates are $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. Wonder Full has no per-token price listed and is restricted to Muse Code, so its cost per task cannot be compared directly from the supplied figures.
On individual tests, Artificial Analysis reported that Wonder Full scored 57% on Terminal-Bench 4.0, up from 33% for Muse Spark 1.3, and 71% on AutomationBench-AA. It also recorded 33% on GDP.pdf and 33% on CritPt, results the evaluation placed at or near the top of those tests. The source says Wonder Full ranked in the top three on six of the ten evaluations. On GDPval-AA, a test of economically useful knowledge work, its 1,857 Elo score was close to Claude Opus 5.5’s 1,866 and within the reported confidence intervals.
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
Wonder’s Benchmark Gains Meet a High Bill
The results position Meta as a stronger contender in frontier-model evaluations, but access and pricing shape how useful the release may be to customers. Wonder Full’s reported results are available only inside Muse Code; businesses seeking a metered API option instead get Wonder, which scored four points lower. That distinction matters for developers who need a model they can integrate into their own products or compare by usage cost.
Artificial Analysis estimated that one Intelligence Index task costs $4.52 with Muse Wonder. In the same comparison, GPT-6.1 Sol scored 52 at $0.72 per task, Gemini 4 Argon scored 53 at $1.99, and Claude Sonnet 5.5 scored 56 at $5.46. These are benchmark-based estimates, not universal prices for every workload; actual bills depend on usage. Still, the figures suggest that Wonder’s strong score does not, by itself, make it a cost-effective choice.
The evaluation also points to a practical reliability issue. It gave Wonder a 27% hallucination rate and Wonder Full 23%, compared with 33% for Muse Spark 1.3. That is a reported improvement, but the figures leave room for errors. In workflows where incorrect outputs can cause harm or financial loss, the benchmark does not remove the need for human review or other verification.
As an affiliate, we earn on qualifying purchases.
From Muse Spark to Two Models
The release follows Muse Spark 1.3 by about six weeks, according to the source report. Artificial Analysis’s figures show a substantial change on several reported measures: Muse Wonder Full rose from 33% to 57% on Terminal-Bench, while Wonder’s score on AutomationBench-AA reached 68%, up from Spark 1.3’s 58%. Wonder Full also received an AA-Omniscience score of 30, compared with 25 for Wonder.
The models are offered through different products, which limits direct comparisons. Wonder is the model customers can call through Meta’s API; Wonder Full is available only in Muse Code. Artificial Analysis’s separate Coding Agents Index rated Muse Code paired with Wonder Full at 68, equal to Claude Code paired with Claude Sonnet 5.5. It rated Claude Code with Opus 5.5 at 66, Muse Code with Wonder at 63, and Muse Code with Spark 1.3 at 54. The index assesses a model together with its coding-agent software, not the model alone.
Artificial Analysis also reported that Wonder used about 7.69 million input tokens per index task, compared with 6.03 million for Spark 1.3. Its output was about 54,000 tokens. The source says input accounted for $3.43, or 76%, of Wonder’s estimated $4.52 task cost. That token use helps explain why its benchmark task estimate can exceed estimates for models with similar scores, although it does not establish what every customer’s workload will cost.
“Meta closed the capability gap. It did not close the price gap.”
— Thorsten Meyer, ThorstenMeyerAI.com
AI model performance evaluation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
API Value and Full’s Cost Remain Open
The published figures do not settle how Muse Wonder performs across customers’ actual applications or whether its API rates will be competitive for particular workloads. Artificial Analysis’s task costs come from its benchmark setup; token consumption, caching, and task design can change bills. The source does not provide an independently verifiable per-task price for Muse Wonder Full, which has no listed per-token rate and is available only in Muse Code.
The material provided also does not include a response from Meta explaining the pricing or the access limits. It is not clear whether Meta plans to make Wonder Full available through an API, publish its usage costs, or change Wonder’s pricing. Benchmark scores and reported error rates are evaluations under specified conditions and should not be read as proof of equivalent results in every deployment.
AI input output token cost calculator
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Watch for Pricing and Wider Access
The next developments to watch are whether Meta expands access to Wonder Full, provides a transparent pricing basis for its use in Muse Code, or adjusts the API rates for Wonder. Further evaluations and customer usage data may also clarify whether the benchmark advantages translate into reliable performance on practical tasks and how much the models cost under different workloads.
For now, buyers can compare the published API price and Artificial Analysis’s task estimate for Wonder with the scores and estimates for competing models, while treating benchmark results as one input to their decision. Meta’s release establishes a stronger benchmark showing; the cost and availability questions remain unresolved in the supplied reporting.
As an affiliate, we earn on qualifying purchases.
Key Questions
What scores did the two Muse Wonder models receive?
Artificial Analysis’s Intelligence Index v4.3 rated Muse Wonder Full at 56 and the API-accessible Muse Wonder at 52. The evaluation was published October 12, 2026.
Where can customers use Muse Wonder Full?
The source report says Muse Wonder Full is available only through Muse Code. It does not list a per-token API price for the model.
How much does Muse Wonder cost?
Meta’s listed API rates are $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. Artificial Analysis estimated $4.52 to run one of its Intelligence Index tasks with Wonder; that benchmark estimate is not a fixed price for every use.
Does a score of 56 mean Wonder Full is better than Claude Opus 5.5?
No. In this evaluation, Wonder Full scored 56 and Claude Opus 5.5 scored 58. The source also says Wonder Full tied Claude Sonnet 5.5 at 56 and that their GDPval-AA results were within the reported confidence intervals.
What remains unknown about the release?
The supplied report does not establish whether Wonder Full will become available through Meta’s API, what its cost per task would be, or whether Meta plans to change Wonder’s pricing. Performance and costs on individual customer workloads may differ from benchmark estimates.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
