🔍 Read the full analysis: Muse Wonder Full And Muse Wonder Charge Into AI on ThorstenMeyerAI.com
Get smart everyday buys 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 AI models evaluated by Artificial Analysis on October 12, 2026. Wonder Full scored 56 on the Intelligence Index, while API-available Wonder scored 52; the reported cost for one index task with Wonder was $4.52, far above some similarly rated competitors.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that scored 52 and 56, respectively, on Artificial Analysis’ Intelligence Index v4.3. The results put Wonder Full near the top of the evaluated models, while the API-available Wonder’s reported $4.52 cost per index task was substantially higher than that of several models with similar scores, raising a practical question for customers weighing capability against cost.
Artificial Analysis published its evaluation on October 12, 2026, according to the source report. Muse Wonder Full scored 56, eight points above Muse Spark 1.3’s 48, and improved on all ten evaluations in the index. Of 27 configurations compared in the report, Claude Opus 5.5 in max mode scored higher, at 58, while Claude Sonnet 5.5 in max mode also scored 56. Muse Wonder scored 52, matching GPT-6.1 Sol in max mode.
The models differ in access and pricing. Muse Wonder is available through Meta’s API and Muse Code, its coding agent, at list prices of $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Muse Wonder Full is available only to Muse Code users and has no per-token price, so its per-task cost cannot be calculated from public pricing. Both models have a one-million-token context window, accept text, images, speech or audio, and video as input, and return text.
The report describes gains across several benchmarks for Wonder Full. It scored 57% on Terminal-Bench 4.0, compared with 33% for Muse Spark 1.3, and ranked first on GDP.pdf and CritPt, at 33% on each. Its GDPval-AA score was 1,857 Elo, between Claude Opus 5.5 at 1,866 and Claude Sonnet 5.5 at 1,838; the report says those results were within confidence intervals. On the Coding Agents Index, the combination of Muse Code and Wonder Full scored 68, tied with Claude Code and Claude Sonnet 5.5. These are evaluations reported by Artificial Analysis, not guarantees of results in every use.
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 Meet Higher API Bills
The release matters because it places Meta’s models closer to leading competitors on independent benchmark measures, while exposing a gap between the stronger model’s performance and the model customers can buy through an API. Wonder Full scored 56, but is restricted to Muse Code; API customers get Muse Wonder, at 52. That distinction makes it harder for developers to access the highest-scoring model on a pay-per-token basis or compare its operating cost directly with competitors.
The source report puts the cost of one Intelligence Index task with Muse Wonder at $4.52. It gives $0.72 for GPT-6.1 Sol at the same score of 52, $1.99 for Gemini 4 Argon at 53, and $5.46 for Claude Sonnet 5.5 at 56. These figures are tied to the report’s task and evaluation setup; they should not be read as a universal price for all workloads. Still, they make API cost a central part of assessing whether Wonder’s performance is useful for a particular application.
Reliability is another consideration. The report gives Muse Wonder a hallucination rate of 27% and Wonder Full 23%, down from 33% for Muse Spark 1.3. It also reports 15% for Gemini 4 Argon. Those benchmark rates do not predict the error rate in every deployment, but they underline why users handling consequential work may need checks on model outputs rather than relying on fluent answers alone.
As an affiliate, we earn on qualifying purchases.
From Spark to Two Wonder Models
The release came six weeks after Muse Spark 1.3, according to the source report. Meta’s new pair separates access to its strongest reported results from access to a priced API product: Wonder Full is available inside Muse Code, while Wonder is offered through both Muse Code and the API. That split is relevant to companies deciding whether they can use the model in existing software or need to work within Meta’s coding-agent environment.
Artificial Analysis’ Intelligence Index combines results from multiple evaluations, while its Coding Agents Index scores a model together with its software harness. The latter distinction matters: the report’s 68 for Muse Code plus Wonder Full is a score for that model-and-harness pairing. It does not show how Wonder Full would perform inside Claude Code, or establish that it would outperform another model in a different setup.
The report also attributes some of Wonder’s task cost to high input-token use. It says Wonder consumes 7.69 million input tokens per index task, compared with 6.03 million for Spark 1.3, while Wonder Full uses 6.44 million. For Wonder, the report assigns $3.43, or 76% of the $4.52 task cost, to input. The remaining cost breakdown is truncated in the supplied source material, so it cannot be fully described here.
“The difference is the bill.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
Costs and Real-World Performance
The reported benchmark results do not establish how the models will perform on every customer’s tasks. The source material does not provide full testing methodology, confidence intervals for every score, or a complete breakdown of the $4.52 task cost. It also does not provide per-task pricing for Muse Wonder Full, which is available only through Muse Code.
It remains unclear whether Meta will offer Wonder Full through its API, publish a way to estimate its cost, or change Wonder’s pricing. The figures for token use and task costs come from the report’s Intelligence Index setup; workloads with different prompts, context lengths or output needs may produce different bills. The source also reports lower hallucination rates than Muse Spark 1.3, but that improvement does not remove the possibility of errors.
As an affiliate, we earn on qualifying purchases.
API Access and Pricing Questions
The immediate next step for prospective users is to test Muse Wonder against their own workloads and measure both quality and token consumption. Developers considering the API can compare the model’s performance with alternatives while accounting for the report’s finding that input tokens make up most of Wonder’s tested task cost. For consequential uses, independent verification remains prudent given the reported hallucination rate.
Meta’s next moves on Wonder Full access and pricing are not specified in the source material. Any expansion beyond Muse Code, a published cost structure, or revised API prices would give customers a clearer basis for comparing the two models with competing services. Until then, the strongest reported benchmark score and the clearest public API price apply to different versions of Meta’s release.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are Muse Wonder and Muse Wonder Full?
They are two proprietary Meta AI models. Both accept text, images, speech or audio, and video as input, return text, and have a one-million-token context window, according to the source report.
Which model can developers access through Meta’s API?
Muse Wonder is available through Meta’s API and inside Muse Code. Muse Wonder Full is available only to Muse Code users, based on the supplied report.
How did the models score?
Artificial Analysis’ Intelligence Index v4.3 gave Muse Wonder Full a score of 56 and Muse Wonder a score of 52. The report says Wonder Full improved on all ten evaluations in the index.
How much does Muse Wonder cost?
Meta’s reported list prices are $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. The source report estimates $4.52 for one Intelligence Index task; that is a benchmark-specific figure, not a fixed cost for every use.
Is Muse Wonder Full available at a public per-token price?
No per-token price is provided in the source material. The report says Wonder Full is available only through Muse Code, so its cost per task cannot be calculated from the listed API prices.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
