🔍 Read the full analysis: Meet Muse Wonder Full, The New Name On AI’s Frontier on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models with a 1-million-token context window. Artificial Analysis rated Muse Wonder Full at 56 on its Intelligence Index, while the API-available Muse Wonder scored 52; its estimated benchmark-task cost was $4.52, compared with $0.72 for GPT-6.1 Sol at the same score. The Full model is available only through Meta’s Muse Code product, and its per-task cost is not public.
Meta has released Muse Wonder and Muse Wonder Full, two new proprietary AI models, with an independent evaluation placing the Full version among the leading models tested. Artificial Analysis gave Muse Wonder Full a score of 56 on its Intelligence Index v4.3, while the version available through Meta’s API scored 52 and had an estimated cost of $4.52 per benchmark task.
Artificial Analysis published its assessment on October 12, 2026. Muse Wonder Full scored eight points above Meta’s Muse Spark 1.3, which received 48, and improved on all ten evaluations included in the index, according to the assessment. Of 27 configurations compared in the source report, only Claude Opus 5.5 in its max configuration scored higher, at 58; Claude Sonnet 5.5 in max configuration also scored 56.
The two Meta models share a 1-million-token context window and accept text, images, audio and video as input, returning text. Their access differs: Muse Wonder is available through Meta’s API and in Muse Code, while Muse Wonder Full is available only to Muse Code users. Meta’s list price for Muse Wonder is $4 per million input tokens and $20 per million output tokens, with cached input listed at $0.20 per million tokens. The source report says Muse Wonder Full has no per-token price, leaving its cost per task unknown.
On individual tests, Artificial Analysis rated Muse Wonder Full at 57% on Terminal-Bench 4.0, up from 33% for Muse Spark 1.3, and 1,857 Elo on GDPval-AA, an evaluation of economically useful knowledge work. It also recorded a 23% hallucination rate for Wonder Full, down from 33% for Spark 1.3. Muse Wonder, the API model, scored 52% on Terminal-Bench and had a reported hallucination rate of 27%. These are benchmark results, not guarantees of performance in a particular deployment.
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, Separate Access Tiers
The release gives Meta a strong result in independent testing, but the distinction between the two models matters to customers. The highest-scoring version, Muse Wonder Full, is restricted to Muse Code, while developers seeking API access get Muse Wonder, which scored 52 on the same Intelligence Index. That makes the headline score unavailable to buyers who want to use the model directly through an API.
The cost comparison in the source report also raises a practical question for teams choosing a model. Artificial Analysis estimated that one Intelligence Index task costs $4.52 with Muse Wonder, compared with $0.72 for GPT-6.1 Sol, which also scored 52. The estimate depends on the evaluation’s workload and token use; it is not a fixed price for every user task. Still, it signals that benchmark strength alone does not settle the value question, particularly for repeated or high-volume use.
The report attributes much of Muse Wonder’s estimated bill to input processing: the model consumed 7.69 million input tokens per index task, with input accounting for $3.43, or 76%, of the $4.52 estimate. The figures suggest that performance and efficiency should be considered together. Customers will need their own workload tests to determine whether the results justify the cost.
As an affiliate, we earn on qualifying purchases.
Six Weeks After Muse Spark
The new models follow Muse Spark 1.3 by about six weeks. In the Artificial Analysis comparison cited by the source report, Spark scored 48 on the Intelligence Index and cost an estimated $1.60 per task. Muse Wonder’s score of 52 is higher, but its estimated task cost is also higher; the report lists Muse Wonder’s input and output prices at $4 and $20 per million tokens, compared with Spark’s $1.25 and $4.25.
Artificial Analysis also evaluates coding systems as combinations of a model and its software harness. Its Coding Agents Index gave Muse Code with Muse Wonder Full a score of 68, matching Claude Code with Claude Sonnet 5.5 in the reported comparison. Muse Code with the API-level Muse Wonder scored 63. Because these ratings cover the full model-and-harness setup, they do not establish how the model would perform in a different coding environment.
Other results were more mixed. Muse Wonder Full scored 30 on AA-Omniscience, below several named competitors in the source report, while Muse Wonder scored 25. Artificial Analysis reported a 23% hallucination rate for Full and 27% for Wonder; the report also cites a 15% rate for Gemini 4 Argon. Such benchmark measures can help compare systems under a shared test, but they do not establish how often a model will make errors in every real-world use.
“Meta closed the capability gap. It did not close the price gap.”
— Thorsten Meyer, ThorstenMeyerAI.com
Full Model Pricing Remains Unknown
The source material does not provide per-token pricing or a cost per task for Muse Wonder Full, which is accessible only within Muse Code. It is therefore not possible from these figures to compare the Full model’s operating cost directly with the API model or rival systems.
The reported $4.52 figure is an Artificial Analysis estimate for one Intelligence Index task, not a universal charge for an ordinary prompt. Actual costs can vary with input length, output length, caching and workload. The material also does not provide details about independent replication, confidence intervals for every benchmark result, or how the models perform across specific customer applications. Benchmark scores and hallucination rates should be treated as test results rather than promises about individual outcomes.
Watch Access and Real-World Costs
The next practical signals will be whether Meta expands access to Muse Wonder Full, publishes a pricing structure, or makes the model available through its API. None of those steps is confirmed in the source material. Until then, organizations evaluating the release will have to distinguish between the performance of the API model they can buy and the stronger model available inside Muse Code.
For developers considering Muse Wonder, the immediate next step is to test it against their own tasks and token volumes, comparing both output quality and total input costs with alternatives. Further independent evaluations and customer experience reports could clarify whether the benchmark results translate into useful gains outside the test suite. Meta has not announced a further model release date in the material provided.
Key Questions
What are Meta’s new Muse models?
Muse Wonder and Muse Wonder Full are proprietary Meta models released about six weeks after Muse Spark 1.3. Both have a reported 1-million-token context window and accept text, image, audio and video inputs.
How did Muse Wonder Full score?
Artificial Analysis gave Muse Wonder Full a score of 56 on its Intelligence Index v4.3, eight points above Muse Spark 1.3. The source report says Claude Opus 5.5 scored 58 and Claude Sonnet 5.5 also scored 56.
Can developers access Muse Wonder Full through an API?
The source material says Muse Wonder Full is available only to Muse Code users. The API-accessible model is Muse Wonder, which scored 52 on the Intelligence Index.
How much does Muse Wonder cost to run?
Meta’s listed API rates for Muse Wonder are $4 per million input tokens and $20 per million output tokens, with cached input at $0.20 per million tokens. Artificial Analysis estimated $4.52 for one Intelligence Index task, but actual costs depend on usage; no per-token price is provided for Muse Wonder Full.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
