🔍 Read the full analysis: Muse Wonder Full Enters The AI Race, With Muse Wonder Charging Too on ThorstenMeyerAI.com
Get hardware and tech essentials 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
Meta has released Muse Wonder and Muse Wonder Full, two proprietary models evaluated by Artificial Analysis. Wonder Full scored 56 on its Intelligence Index, while the API model scored 52; benchmark task estimates put Wonder at $4.52, compared with $0.72 for GPT-6.1 Sol at the same score.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that received markedly different access and pricing arrangements. In an evaluation published October 12, Artificial Analysis gave Wonder Full a score of 56 on its Intelligence Index, while Wonder, the version offered through Meta’s API, scored 52 and was estimated to cost $4.52 per benchmark task.
Artificial Analysis said Wonder Full improved on all ten evaluations in its Intelligence Index v4.3 and placed in the top three on six. Its score of 56 was below Claude Opus 5.5’s 58 and tied Claude Sonnet 5.5’s score. The evaluation is an independent assessment cited by the source report, not a measure of how every user will experience the models.
The two models share a one-million-token context window and accept text, images, audio or speech, and video as input, returning text. Meta makes Wonder available through its API and inside Muse Code, its coding agent. Wonder Full is available only to Muse Code users and has no published per-token price, so its cost per task cannot be compared from the supplied figures.
Meta’s listed Wonder API rates are $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. Artificial Analysis estimated that one Intelligence Index task cost $4.52 on Wonder. Its report attributes 76% of that total, or $3.43, to input tokens; Wonder processed about 7.69 million input tokens per task in the evaluation.
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 Score Comes With a High API Bill
The release puts Meta’s progress and its commercial positioning in the same frame. Wonder Full’s score places it close to leading models on this benchmark, while the API-accessible Wonder scored 52, the same as GPT-6.1 Sol in the comparison. Yet Artificial Analysis estimated a task cost of $4.52 for Wonder versus $0.72 for Sol. These are benchmark-specific cost estimates, not a promise of what any particular customer will spend.
That difference matters to teams comparing models for repeated workloads: a small change in benchmark score can look different when paired with a large difference in estimated task cost. The source report also says Wonder’s high input-token use is a major driver of the bill. Actual costs will depend on prompts, usage patterns, and the applicable API pricing.
For coding, the reported Coding Agents Index gives Muse Code with Wonder Full a score of 68, tied with Claude Code using Sonnet 5.5. That index measures the model and its software harness together. It does not establish that Wonder Full would achieve the same result in another coding environment or that it outperforms Claude Opus 5.5 across uses.
AI language model API subscription
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Six Weeks After Muse Spark 1.3
The release follows Muse Spark 1.3 by six weeks, according to the source report. Artificial Analysis scored Spark 1.3 at 48 on the Intelligence Index, compared with 56 for Wonder Full and 52 for Wonder. The report also cites improvements on individual measures: Wonder Full scored 57% on Terminal-Bench 4.0, up from 33% for Spark 1.3, while Wonder scored 52%.
Those benchmark results cover different abilities, including shell-based task completion, knowledge work, and accuracy. The source report says Wonder Full scored 1,857 Elo on GDPval-AA, close to Claude Opus 5.5 at 1,866 and Claude Sonnet 5.5 at 1,838; it says the results are within the models’ confidence intervals. Such scores offer a point of comparison, but do not by themselves establish performance in a particular company’s workflow.
The report also gives hallucination rates of 23% for Wonder Full and 27% for Wonder, down from 33% for Spark 1.3 on the cited measure. It says Gemini 4 Argon recorded 15% among models scoring at least 45 on the Intelligence Index. These results reinforce that higher overall benchmark scores do not remove the risk of incorrect answers.
“That gap is the story.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
Full Model Costs Remain Unpublished
The supplied material does not include Meta’s own explanation for Wonder’s pricing or how it set the rates. It also provides no per-token price for Muse Wonder Full, which is limited to Muse Code users. That leaves its operating cost difficult to compare with API models.
The $4.52 figure is Artificial Analysis’s estimate for one benchmark task, not a fixed charge for every real-world request. The report excerpt ends during its token-cost breakdown after identifying input as $3.43, or 76% of Wonder’s task estimate; it does not provide the remaining breakdown. The source material also does not detail the evaluation’s full methodology or the uncertainty ranges for every listed score.
It is not clear from the information provided how widely the models are available across regions, whether access limits apply, or whether Meta plans to change prices. Benchmark scores and hallucination rates also cannot establish reliability for a specific use case. Users handling consequential work would need to test the models against their own tasks and verify outputs.
As an affiliate, we earn on qualifying purchases.
Pricing and Access Need Clarification
The immediate next questions are whether Meta will publish a price for Wonder Full and whether it will explain or revise Wonder’s API rates. The source material does not identify a date for either development or give a schedule for wider access.
For now, customers can compare the published API rates and independent benchmark estimates, while recognizing that task costs vary with token use. Further evaluations, pricing details, and information about availability would help clarify whether Wonder’s performance justifies its estimated cost for particular workloads.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Meta release?
Meta released Muse Wonder and Muse Wonder Full. Both are proprietary models with a one-million-token context window, according to the source report.
How did the models score?
Artificial Analysis scored Wonder Full at 56 and Wonder at 52 on its Intelligence Index v4.3. The scores summarize performance across the index’s evaluations and do not guarantee results for an individual task.
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 a cost of $4.52 for one Intelligence Index task; actual charges depend on usage.
Can customers buy Muse Wonder Full through the API?
Not according to the source report. Wonder Full is available only to Muse Code users, and no per-token price is provided in the material.
Do the benchmark results mean the models cannot make mistakes?
No. The report cites hallucination rates of 23% for Wonder Full and 27% for Wonder on its measure. Those figures indicate that errors remain possible; benchmark results do not replace verification for important work.
Source: ThorstenMeyerAI.com
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
