🔍 Read the full analysis: Muse Wonder Full Steps Into AI’s Frontier And Muse Wonder Charges Again 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
Meta has released Muse Wonder and Muse Wonder Full, with independent testing by Artificial Analysis showing substantially higher benchmark scores than Muse Spark 1.3. The stronger Full model is available only through Muse Code, while the API-available Muse Wonder costs $4.52 per Intelligence Index task in the evaluator’s comparison—more than several models with similar or higher scores.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that scored above Muse Spark 1.3 in Artificial Analysis testing, but the stronger model is restricted to Meta’s coding agent and the API model’s evaluated task cost is high relative to several competitors. Artificial Analysis reported on October 12 that Muse Wonder Full scored 56 on its Intelligence Index v4.3, while the API-available Muse Wonder scored 52.
Artificial Analysis said Muse Wonder Full improved on all ten evaluations in its Intelligence Index and placed in the top three on six. Its score of 56 was eight points above Muse Spark 1.3’s 48. In the evaluator’s comparison of 27 configurations, Claude Opus 5.5 at its maximum setting scored 58, while Claude Sonnet 5.5 at maximum tied Muse Wonder Full at 56. These are benchmark results, not guarantees of performance in every task.
The models differ in access. Muse Wonder is available through Meta’s API and inside Muse Code, while Muse Wonder Full is available only to Muse Code users. Both have a reported one-million-token context window and accept text, images, audio and video as input, returning text. Meta’s stated list prices for Muse Wonder 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 listed, so its cost per task cannot be calculated from the source material.
Artificial Analysis reported that Muse Wonder Full scored 57% on Terminal-Bench 4.0, compared with 33% for Muse Spark 1.3, and placed third on AutomationBench-AA with 71%. Its GDPval-AA score was 1,857 Elo, close to Claude Opus 5.5’s 1,866 and Claude Sonnet 5.5’s 1,838; the source says those results are within confidence intervals. For the API model, Artificial Analysis reported a 52% Terminal-Bench result and 68% on AutomationBench-AA, up from 58% for Spark 1.3.
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
Benchmark Gains Meet API Costs
The release puts Meta closer to leading proprietary models on several measured tasks, particularly coding and economically oriented knowledge work. Artificial Analysis’s Coding Agents Index gave Muse Code with Muse Wonder Full a score of 68, tying its Claude Code plus Claude Sonnet 5.5 configuration. The index evaluates the model together with its software harness, so it does not establish how Wonder Full would perform inside a different coding product.
For customers choosing an API, the cost comparison is a separate consideration from raw benchmark performance. Artificial Analysis estimated that one Intelligence Index task cost $4.52 with Muse Wonder, versus $0.72 for GPT-6.1 Sol at the same score of 52 and $1.99 for Gemini 4 Argon, which scored 53. These estimates are specific to the evaluator’s task and method; they do not represent a universal price for every workload. Still, they raise a practical question for developers: whether the model’s measured gains justify its API bill and token use in their own applications.
The comparison also matters because Muse Wonder Full cannot be purchased through the API on the reported terms. Developers can test the higher-scoring model in Muse Code, but the source provides no per-token price or independent task-cost estimate for it. That limits direct cost comparisons between Meta’s two releases and makes the API model’s price-performance profile the more visible part of the launch.
As an affiliate, we earn on qualifying purchases.
From Spark to Wonder
The two models arrived about six weeks after Muse Spark 1.3, according to the source article. Artificial Analysis’s reported Intelligence Index scores were 48 for Spark 1.3, 52 for Muse Wonder and 56 for Muse Wonder Full. The spread indicates a marked change in that particular evaluation, though benchmark scores alone do not establish that every user or business task will improve by the same amount.
Price lists provide one explanation for the scrutiny around the API release. The source reports that Spark 1.3 was priced at $1.25 per million input tokens and $4.25 per million output tokens. Muse Wonder’s listed rates are $4 and $20, respectively, matching the reported list prices for Claude Opus 5.5. It also reports that Claude Sonnet 5.5, Gemini 4 Argon and GPT-6.1 Sol each list at $2 per million input tokens and $10 per million output tokens.
Artificial Analysis’s task-cost estimate reflects the tokens used in its benchmark run as well as list prices. It reported that Muse Wonder consumed 7.69 million input tokens per index task, compared with 6.03 million for Spark 1.3, and generated about 54,000 output tokens. The source says input accounted for $3.43, or 76%, of the $4.52 estimated task cost. Those figures describe the evaluator’s test setup, not a fixed bill for other prompts or workloads.
“The bill is in the input, not the output.”
— Thorsten Meyer
As an affiliate, we earn on qualifying purchases.
Access and Cost Questions Remain
The source material does not provide a per-token price for Muse Wonder Full, which is limited to Muse Code, or an equivalent independent task-cost estimate. It is therefore not possible to determine from these figures whether the stronger model is less or more expensive to use for comparable work.
The benchmark and cost numbers come from Artificial Analysis evaluations and depend on the tasks, configurations and token volumes used. The source does not provide enough detail here to determine how results would vary across different prompts, software environments or customer workloads. Nor does it establish that benchmark rankings will translate into better outcomes for every use case.
Artificial Analysis reported hallucination rates of 23% for Muse Wonder Full and 27% for Muse Wonder, down from 33% for Spark 1.3, while identifying lower rates for some competing systems. The source cautions that confident errors remain a concern. The precise reliability a user can expect in a specific deployment is not established by these aggregate results, so organizations still need to test outputs and decide what verification their work requires.
As an affiliate, we earn on qualifying purchases.
Real-World Pricing Will Matter
The next practical test is how the models perform and bill in customer workloads, beyond Artificial Analysis’s benchmark tasks. Developers can access Muse Wonder through Meta’s API and compare its performance, token consumption and total cost against alternatives using their own prompts. Meta’s listed rates provide a basis for estimating usage, but actual spending will depend on input volume, output length, caching and task design.
For Muse Wonder Full, further clarity would require information about its availability beyond Muse Code and how its use is priced. The source does not report a timetable for broader access or a change to the current access arrangement. Until Meta provides those details, comparisons involving the full model will remain focused on benchmark results and the Muse Code configuration rather than a public API price.
AI video and audio processing software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Meta release?
Meta released Muse Wonder and Muse Wonder Full. Artificial Analysis reported Intelligence Index v4.3 scores of 52 for Muse Wonder and 56 for Muse Wonder Full.
Which model is available through Meta’s API?
Muse Wonder is available through Meta’s API and inside Muse Code. The source says Muse Wonder Full is available only to Muse Code users.
How much does Muse Wonder cost?
The reported list prices are $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Artificial Analysis estimated $4.52 for one of its Intelligence Index tasks; that figure depends on its test’s token use.
How did Muse Wonder Full compare with leading models?
Artificial Analysis scored it 56, tying Claude Sonnet 5.5 at maximum settings and trailing Claude Opus 5.5 at 58 in the cited comparison. Its Muse Code configuration also tied a Claude Code configuration at 68 on the Coding Agents Index.
Are the models reliable enough to use without checking their answers?
The source reports lower hallucination rates than for Muse Spark 1.3, but still records rates of 23% for Muse Wonder Full and 27% for Muse Wonder in the cited evaluation. Those results do not establish reliability for every task; users should test outputs and apply verification appropriate to their work.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
