🔍 Read the full analysis: Can Muse Wonder Full Find Its Place On The AI Frontier? 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
Artificial Analysis gave Meta’s Muse Wonder Full a score of 56 on its Intelligence Index, tying Claude Sonnet 5.5 and trailing Claude Opus 5.5 by two points. The API-available Muse Wonder scored 52, but the evaluator estimated its cost at $4.52 per index task, compared with $0.72 for GPT-6.1 Sol at the same score. The stronger Full model is available only through Meta’s Muse Code, and its per-task cost is not public.
Meta’s Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index, tying Claude Sonnet 5.5 and coming two points below Claude Opus 5.5, according to an independent evaluation published October 12. The result puts Meta’s latest model among the top performers tested, but the model available through Meta’s API, Muse Wonder, scored 52 and was estimated to cost $4.52 per index task—far more than some similarly scoring competitors.
Artificial Analysis reported that Muse Wonder Full improved on all 10 evaluations in its Intelligence Index v4.3 compared with Muse Spark 1.3. It placed in the top three on six evaluations. Its Terminal-Bench 4.0 score rose to 57% from 33%, while its GDPval-AA result was 1,857 Elo, between Claude Opus 5.5 at 1,866 and Claude Sonnet 5.5 at 1,838. The evaluator said those GDPval-AA results fall within the models’ confidence intervals.
The two releases differ in access. Muse Wonder is available through Meta’s API and inside Muse Code, Meta’s coding agent. Its listed rates are $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; it has no published per-token price, so the cost of running it on the same tasks cannot be calculated from the supplied figures.
Artificial Analysis gave Muse Wonder a score of 52, equal to GPT-6.1 Sol (max), and reported a 52% Terminal-Bench score. Its estimated index-task cost was $4.52, versus $0.72 for Sol. The evaluator also reported that Wonder uses 7.69 million input tokens per index task, compared with 6.03 million for Muse Spark 1.3, and estimated that input accounted for 76% of Wonder’s task cost.
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
Frontier Scores, Different Price Tiers
The results point to a split between Meta’s strongest model and its commercially accessible one. Muse Wonder Full’s score places it close to leading models in this evaluation, but its restricted availability means customers cannot choose it through a standard API price-per-token offering. Meanwhile, buyers comparing API models face a substantial price difference: Artificial Analysis estimated Muse Wonder’s task cost at 6.3 times GPT-6.1 Sol’s, even though both scored 52.
Those comparisons matter for teams weighing model quality against operating costs. A benchmark score alone does not establish which model will be cheaper or more effective for a particular workload, but the reported cost gap gives prospective users a reason to test their own prompts and usage patterns before committing. The evaluation’s token figures suggest that input volume, not just listed token rates, contributes heavily to the estimated bill.
The coding-agent result adds another dimension. Artificial Analysis scored Muse Code paired with Muse Wonder Full at 68 on its Coding Agents Index, matching Claude Code paired with Claude Sonnet 5.5. That index assesses the model and its software harness together; it is not a direct comparison of the models running in identical environments.
AI model performance evaluation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Two Muse Models Offer
Meta released Muse Wonder and Muse Wonder Full six weeks after Muse Spark 1.3, according to ThorstenMeyerAI.com. Both are proprietary models with a one-million-token context window. They accept text, images, speech or audio, and video as input, and return text. The source report identifies different access routes: API and Muse Code for Wonder, and Muse Code only for Wonder Full.
The independent evaluation reported improvements beyond coding tasks. Muse Wonder Full scored 33% on GDP.pdf and 33% on CritPt, leading the compared models on those evaluations, whose nearest rivals scored 31% to 32%. It scored 71% on AutomationBench-AA and 63% on SciCode, ranking third on each. Its AA-Omniscience score was 30, up five points from Muse Spark 1.3, while its reported hallucination rate fell from 33% to 23%.
Muse Wonder also improved on its predecessor on selected measures: it scored 68% on AutomationBench-AA, compared with 58% for Muse Spark 1.3, and had a reported hallucination rate of 27%. Yet both new models remained below several rivals on AA-Omniscience, a measure that rewards correct answers and penalizes confident errors. Artificial Analysis listed scores of 46 for Claude Opus 5.5 and 43 for GPT-6 Astra, compared with 30 for Wonder Full and 25 for Wonder.
“The gap is the story.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
Costs and Real-World Performance
The reported costs are Artificial Analysis estimates for its Intelligence Index tasks, not a guarantee of what a customer will spend on a different workload. The supplied material does not give the complete cost breakdown beyond the reported $3.43 attributed to input and its 76% share, so the remaining components cannot be independently reconstructed here. Actual bills may vary with prompt length, caching, output volume and the task being run.
It is also unclear how Muse Wonder Full compares with other models on a like-for-like API basis: it has no public per-token price and is not offered through Meta’s API in the source report. The coding-agent score combines a model with its particular harness, so it does not establish that Wonder Full would score the same way inside Claude Code or another environment. Benchmark results also do not establish performance across every customer’s use case.
Artificial Analysis reported lower hallucination rates for the new Muse models than for Muse Spark 1.3, but the figures remain higher than Gemini 4 Argon’s reported 15% among models scoring at least 45 on the Intelligence Index. The source recommends retaining verification for workflows where confident errors carry costs; the evaluation alone does not show how often such errors would cause harm in a specific deployment.
As an affiliate, we earn on qualifying purchases.
API Pricing and Independent Tests
The next practical test is whether Meta changes Muse Wonder’s pricing, token use or API offering, or makes Muse Wonder Full available beyond Muse Code. The supplied report does not announce any planned price adjustment or broader rollout. Users considering the models can compare them on their own tasks and monitor additional independent evaluations rather than relying on one index score.
Further evidence will be needed to clarify how the models perform across real deployments, what costs Wonder Full incurs inside Muse Code, and whether its benchmark results translate into consistent gains for coding and knowledge-work tasks. Until Meta publishes more detail, the distinction remains: the higher-scoring model has restricted access, while the API model’s estimated benchmark-task cost is markedly above that of some rivals with similar scores.
As an affiliate, we earn on qualifying purchases.
Key Questions
How did Muse Wonder Full score?
Artificial Analysis gave Muse Wonder Full a 56 on its Intelligence Index v4.3. It tied Claude Sonnet 5.5 and scored two points below Claude Opus 5.5 in the comparison cited by the report.
Can customers access Muse Wonder Full through Meta’s API?
No. The source report says Muse Wonder Full is available only through Muse Code. Muse Wonder is the model available through Meta’s API and Muse Code.
Why is Muse Wonder’s estimated task cost a concern?
Artificial Analysis estimated a cost of $4.52 per Intelligence Index task for Muse Wonder, compared with $0.72 for GPT-6.1 Sol, which also scored 52. These are benchmark-task estimates; costs for other workloads can differ.
Does Muse Wonder Full lead every benchmark?
No. It ranked in the top three on six of the ten Intelligence Index evaluations, and its scores varied by task. For example, Artificial Analysis reported an AA-Omniscience score of 30, below several listed competitors.
What remains unknown about Muse Wonder Full’s price?
The source report gives no per-token price or comparable per-task cost for Wonder Full. Because access is limited to Muse Code, the supplied figures do not show what it would cost to run through a public API.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
