🔍 Read the full analysis: Muse Wonder Full Makes Its AI Frontier Move As Muse Wonder Charges on ThorstenMeyerAI.com
Get the latest gadgets 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 with a reported one-million-token context window. Artificial Analysis scores Wonder Full at 56 and the API-available Wonder at 52, while estimating a $4.52 cost per Intelligence Index task for Wonder—more than several similarly scoring competitors. The evaluation is independent, but task-cost figures are benchmark estimates, not a universal price for every use.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that mark a reported jump over its Muse Spark 1.3 system. In an evaluation published October 12 by Artificial Analysis, Wonder Full scored 56 on the Intelligence Index, while Wonder—the version available through Meta’s API—scored 52; the evaluation estimated that one index task with Wonder costs $4.52, more than several competing models with similar scores.
Artificial Analysis said Wonder Full gained eight points over Muse Spark 1.3’s score of 48 and improved on all ten evaluations in its Intelligence Index v4.3. Among 27 configurations compared in the source report, only Claude Opus 5.5 at its maximum setting scored higher, at 58; Claude Sonnet 5.5 at maximum tied Wonder Full at 56. These are results from the evaluator’s benchmark, not a guarantee of performance on every user task.
The report highlighted a rise on Terminal-Bench 4.0, a test of work completed in a command-line environment: Wonder Full scored 57%, compared with 33% for Muse Spark 1.3. It also placed near Anthropic models on GDPval-AA, an evaluation of economically useful knowledge work. Artificial Analysis cautioned that comparisons between a model and a model-plus-tooling setup need to be read in context.
Muse Wonder is the API model and is also available inside Meta’s Muse Code coding agent. The listed price is $4 per million input tokens and $20 per million output tokens, with cached input listed at $0.20 per million tokens. Wonder Full is available only to Muse Code users, according to the source report, and has no per-token price disclosed there. Both models reportedly accept text, images, audio and video, return text, and have a one-million-token context window.
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
The Cost of Matching Rivals
The release gives developers and businesses a new option from Meta, but the gap between benchmark score and estimated expense may shape whether they try it at scale. Artificial Analysis estimated $4.52 per Intelligence Index task for Muse Wonder, compared with $0.72 for GPT-6.1 Sol at the same score of 52, and $1.99 for Gemini 4 Argon, which scored 53. The report’s estimates apply to the evaluator’s task mix; actual bills will depend on usage and token volume.
That distinction matters because Wonder’s API list prices are higher than those cited for several rivals, and the evaluation found unusually high input-token use. The report estimated about 7.69 million input tokens per task for Wonder, versus 6.03 million for Spark 1.3. Its calculation attributed 76% of Wonder’s estimated task cost to input tokens. This suggests that token consumption, as well as per-token rates, contributes to the cost comparison.
Wonder Full’s results may be relevant to organizations using Meta’s coding agent, but its limited availability and undisclosed per-token cost make direct price comparisons difficult. The benchmark numbers are a signal to test the models, not a substitute for measuring accuracy, latency and cost on an organization’s own work.
As an affiliate, we earn on qualifying purchases.
From Spark to Two Models
The source report places the release about six weeks after Muse Spark 1.3. Meta’s new lineup separates access to the higher-scoring model from access to the model priced for API use: Wonder Full is restricted to Muse Code, while Wonder is offered through Meta’s API and that coding agent. That division means users cannot assume the top Intelligence Index result is available as a generally priced API service.
Artificial Analysis also evaluated coding-agent combinations, which combine the model with its software harness. Muse Code with Wonder Full scored 68, matching Claude Code with Sonnet 5.5 in the Coding Agents Index; Claude Code with Opus 5.5 scored 66 in that comparison. These figures describe the tested combinations. They do not establish how Wonder Full would perform inside Claude Code or how the models compare in other tools.
The report also found improvement in its hallucination measure: Wonder Full scored 30 on AA-Omniscience and Wonder scored 25, while the source gives hallucination rates of 23% and 27%, respectively. Spark 1.3’s reported rate was 33%. The measure is specific to the evaluator, and the results do not remove the need to verify outputs in work where errors carry consequences.
“The difference is the bill.”
— Thorsten Meyer, ThorstenMeyerAI.com
Access, Pricing and Real-World Results
The source material does not provide a release date beyond saying the models shipped about six weeks after Muse Spark 1.3, nor does it spell out regional availability or API rate limits. It also gives no per-token price for Wonder Full, which is limited to Muse Code. The cost comparisons are based on Artificial Analysis’s index-task methodology; the supplied material does not fully describe how representative that task is of typical customer workloads.
Benchmark rankings cannot establish how the models will perform across different prompts, tools, or production settings. The report’s hallucination figures are evaluation-specific, and the source does not provide confidence intervals for every result. The numbers should not be treated as proof that one model will be more reliable or less expensive for every user.
Testing Beyond the Index
The next practical test is whether developers and businesses find Wonder’s performance worth its measured token use and listed API rates on their own workloads. API users can compare results, input and output volume, and costs against alternatives; teams considering Wonder Full will also need clearer access and pricing information to judge its economics.
Meta has not, in the supplied material, announced further availability or pricing details for Wonder Full. Artificial Analysis’ published results provide an initial comparison, but broader adoption, independent evaluations and customer experience will show whether the benchmark gains translate into useful performance outside the index.
Key Questions
What did Meta release?
Meta released Muse Wonder and Muse Wonder Full, proprietary models that the source says support text, images, audio and video input and have a one-million-token context window.
How did the models score?
Artificial Analysis gave Wonder Full a score of 56 and Wonder a score of 52 on its Intelligence Index v4.3. These are benchmark results, not guaranteed outcomes for all tasks.
Which model can developers access through Meta’s API?
Muse Wonder is the model available through Meta’s API, according to the source report. Wonder Full is described as available only to Muse Code users.
Why is Muse Wonder’s estimated task cost drawing attention?
Artificial Analysis estimated a cost of $4.52 per index task for Wonder, compared with $0.72 for GPT-6.1 Sol, which received the same score of 52. The estimate reflects the evaluator’s task and should not be read as a fixed cost for every API call.
What is still unknown?
The supplied report does not state Wonder Full’s per-token price or detail its availability beyond Muse Code. It is also unclear how the benchmark results and task-cost estimates will translate to individual customer workloads.
Source: ThorstenMeyerAI.com
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
