🔍 Read the full analysis: Muse Wonder Full Joins The AI Frontier: What To Know on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Artificial Analysis’s October 12 evaluation gave Meta’s Muse Wonder Full an Intelligence Index score of 56, up from 48 for Muse Spark 1.3. The model is available only through Meta’s Muse Code, while the API-accessible Muse Wonder scored 52 and cost $4.52 per benchmark task in the evaluation—more than six times the reported cost for GPT-6.1 Sol at the same score.
Meta has released two proprietary AI models, Muse Wonder and Muse Wonder Full, and an independent evaluation by Artificial Analysis published October 12 gave the stronger model an Intelligence Index score of 56. That is eight points above Muse Spark 1.3, but the higher-scoring Full version is available only inside Meta’s Muse Code, while the API-accessible Wonder scored 52 and was estimated to cost $4.52 per benchmark task.
Artificial Analysis said Muse Wonder Full improved on all 10 evaluations in its Intelligence Index v4.3. It placed in the top three on six, according to the source report. The evaluation put it at 57% on Terminal-Bench 4.0, compared with 33% for Muse Spark 1.3, and at 1,857 Elo on GDPval-AA v2.1, a measure of economically useful knowledge work. The report said the GDPval-AA results for Muse Wonder Full and Claude Opus 5.5 and Sonnet 5.5 fell within one another’s confidence intervals.
Muse Wonder is the version available through Meta’s API and in 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 has no per-token price and is limited to Muse Code users, so its cost per task cannot be calculated from the information provided. Both models have a reported one-million-token context window, accept text, images, speech or audio, and video as input, and return text.
Artificial Analysis also scored model-and-tool combinations in its Coding Agents Index. Muse Code with Muse Wonder Full scored 68, matching Claude Code with Claude Sonnet 5.5; Claude Code with Opus 5.5 scored 66. The source cautions that this index assesses a model together with its coding harness, so it does not establish how Wonder Full would perform inside Claude Code.
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
Performance Comes With an API Price
The release puts Meta’s progress and its product trade-off in the same frame. Muse Wonder Full’s benchmark results place it among leading models in several tested areas, including coding-agent performance. But businesses and developers who want API access cannot buy that version by the token; they have access to the lower-scoring Muse Wonder instead.
In Artificial Analysis’s reported comparison, one Intelligence Index task cost $4.52 for Muse Wonder and $0.72 for GPT-6.1 Sol, though both models scored 52. Gemini 4 Argon scored 53 at a reported $1.99 per task, while Claude Sonnet 5.5 scored 56 at $5.46. These are benchmark-task cost estimates, not universal prices for every workload. Actual bills depend on usage, and the figures should be read in the evaluation’s stated context.
The comparison may matter to teams choosing models for repeated work: benchmark scores alone do not show the expense of running a system. The report attributes much of Muse Wonder’s task cost to input-token use. It processed 7.69 million input tokens per task, compared with 6.03 million for Muse Spark 1.3 and substantially lower amounts for some named rivals. The evaluation does not, by itself, establish how those costs or results translate to a particular customer’s workload.
As an affiliate, we earn on qualifying purchases.
From Muse Spark to Two Models
The release came about six weeks after Muse Spark 1.3, according to the source report. That earlier model scored 48 on the same Artificial Analysis Intelligence Index version used for the new comparison. Muse Wonder Full’s 56 and Muse Wonder’s 52 show different performance levels within Meta’s new release, as well as different routes to access.
The report describes Muse Wonder as a substantial improvement over its predecessor in some tested tasks. It records a rise on Terminal-Bench from 33% to 52% for the API model and on AutomationBench-AA from 58% to 68%. It also reports a fall in Muse Wonder’s hallucination rate from 33% to 27%, while Muse Wonder Full’s rate was 23%. Those figures come from the evaluation; they do not mean the models never produce incorrect answers.
Artificial Analysis’s Intelligence Index combines multiple evaluations, so its total score is not a direct measure of performance in every use case. The Coding Agents Index uses a different setup, scoring the model and the software harness together. Keeping those measures distinct matters when comparing Meta’s tools with products built by other companies.
“The bill is in the input, not the output.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
Access and Real-World Costs
Muse Wonder Full’s per-task cost is unknown from the source material because it has no listed per-token price and is restricted to Muse Code. The report also does not provide enough information here to determine how the benchmark cost estimates would compare with bills for specific customer workloads, which may use different prompts, context lengths and task mixes.
Benchmark scores and reported hallucination rates describe results under evaluation conditions, not guarantees for production use. The source gives relative results across selected tests but does not establish how the models perform across every industry or task. It also does not include independent verification of the model prices or the underlying token totals beyond its attribution to Artificial Analysis.
As an affiliate, we earn on qualifying purchases.
Pricing and Wider Availability
The next practical developments to watch are whether Meta makes Muse Wonder Full available beyond Muse Code and whether it publishes a price that allows direct cost comparisons. Meta could also change Muse Wonder’s listed rates or token efficiency over time, but the source material does not report any announced pricing changes or expansion plans.
For now, the evaluation offers a benchmark snapshot rather than a final verdict on value. Developers considering the API model can compare its listed rates and reported token consumption with their own expected workloads, while treating benchmark scores as one input rather than a promise of performance. Any decision involving consequential outputs should retain human review or other verification, particularly because the report records hallucinations for both new models.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are Muse Wonder and Muse Wonder Full?
They are two proprietary models released by Meta. Both accept text, images, speech or audio, and video as input and return text, with a reported one-million-token context window.
Which version can developers access through Meta’s API?
Muse Wonder is available through Meta’s API and in Muse Code. Muse Wonder Full is reported as available only to Muse Code users.
How did Muse Wonder Full score?
Artificial Analysis gave it 56 on its Intelligence Index v4.3, eight points above Muse Spark 1.3’s score of 48. The index combines results across multiple evaluations; it does not guarantee performance on a particular task.
How much does Muse Wonder cost?
Meta’s listed rates in the source report 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; that estimate is specific to the benchmark evaluation.
Is Muse Wonder Full cheaper or more expensive to run?
The source does not give a per-token price for Muse Wonder Full, which is limited to Muse Code. Its cost per task cannot be determined from the information provided.
Source: ThorstenMeyerAI.com
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
