🔍 Read the full analysis: Muse Wonder Full Joins The AI Frontier, With Muse Wonder Charging Like Before on ThorstenMeyerAI.com
Get hardware and tech essentials 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 one-million-token context window. Artificial Analysis scored Wonder Full at 56, but it is available only through Muse Code; the API model scored 52 and cost $4.52 per Intelligence Index task in the evaluator’s test, compared with $0.72 for GPT-6.1 Sol at the same score.
Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that scored strongly in an independent evaluation but differ in access and cost. Artificial Analysis gave Wonder Full a score of 56 on its Intelligence Index v4.3, while the API-available Wonder scored 52; the evaluator calculated a cost of $4.52 per benchmark task for Wonder, compared with $0.72 for GPT-6.1 Sol, which also scored 52.
Artificial Analysis published its evaluation on October 12, 2026. Wonder Full’s score of 56 was eight points above Muse Spark 1.3’s 48, and the evaluator reported improvement across all ten assessments in the index. Among 27 model configurations compared in the source, only Claude Opus 5.5 in its max configuration scored higher, at 58; Claude Sonnet 5.5 max also scored 56. These are benchmark results, not a guarantee of performance in every use case.
The stronger model is not available as a metered API product: Wonder Full is limited to Muse Code, Meta’s coding agent, and has no published per-token price in the source material. Muse Wonder is available through Meta’s API and Muse Code. Its listed price is $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Both models accept text, images, speech or audio, and video as input, return text, and have a one-million-token context window.
Wonder Full placed in the top three on six of the ten evaluations, according to Artificial Analysis. On Terminal-Bench 4.0, it scored 57%, up from 33% for Spark 1.3. It also scored 1,857 Elo on GDPval-AA v2.1, a measure of economically useful knowledge work; the source places that result close to Claude Opus 5.5 and Sonnet 5.5, whose scores fall within the reported confidence intervals. Wonder scored 52 on the Intelligence Index, matching GPT-6.1 Sol’s score, but its benchmark-task cost was much higher.
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 Results, Uneven Access and Cost
The release puts Meta closer to leading models on independent benchmarks, particularly in coding and agent-style tasks. Artificial Analysis’s Coding Agents Index gave Muse Code with Wonder Full a score of 68, tied with Claude Code using Sonnet 5.5. That index evaluates the model together with its software harness, so it does not establish how Wonder Full would perform in a different coding environment. It does indicate that Meta’s own coding-agent setup scored alongside a leading competitor configuration in this test.
For customers choosing an API, the central trade-off is performance versus operating cost. Artificial Analysis estimated that one Intelligence Index task cost $4.52 on Wonder. The same task cost $1.99 on Gemini 4 Argon and $0.72 on GPT-6.1 Sol in the comparison, while those models scored 53 and 52 respectively. The figures describe the evaluator’s benchmark workload; actual costs vary with usage and may not predict a customer’s bill.
Reliability remains another consideration. Artificial Analysis put Wonder’s hallucination rate at 27% and Wonder Full’s at 23%, down from 33% for Spark 1.3. The evaluator also reported a 15% rate for Gemini 4 Argon among models scoring at least 45 on the index. Those test results do not establish how often any model will make errors in a particular deployment. High-stakes uses still call for appropriate human review and verification.
As an affiliate, we earn on qualifying purchases.
Six Weeks After Muse Spark 1.3
The launch follows Muse Spark 1.3 by six weeks, according to the source. In Artificial Analysis’s comparison, Spark 1.3 scored 48 on the Intelligence Index. Wonder Full’s 56 represents a marked increase on that measure, while the API model Wonder scored 52. The comparison provides a snapshot of benchmark performance at release, not evidence about how the models will perform after future updates or in different products.
Pricing also changed between the API models. The source lists Spark 1.3 at $1.25 per million input tokens and $4.25 per million output tokens, compared with Wonder’s $4 and $20 rates. Artificial Analysis estimated that Wonder consumed 7.69 million input tokens for an Intelligence Index task, versus 6.03 million for Spark 1.3, and around 54,000 output tokens, versus 60,000 for Spark. The evaluator attributed $3.43, or 76% of Wonder’s $4.52 task cost, to input tokens. This makes input volume a major part of the reported cost difference, alongside the listed rates.
The two new models also have different product roles: Wonder is the metered API offering, while Wonder Full is available inside Muse Code only. That distinction limits direct cost comparisons for Full because the source provides no per-token price or per-task charge for it.
“The bill is in the input, not the output.”
— Thorsten Meyer, ThorstenMeyerAI.com
AI performance evaluation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Full Model Pricing Remains Unknown
The source does not provide a public per-token or per-task price for Muse Wonder Full, so its operating cost cannot be compared directly with API models from the figures supplied. It is also unclear whether Meta plans to offer Full through an API or disclose pricing for its use in Muse Code.
The reported task costs come from Artificial Analysis’s evaluation setup, which used different numbers of input and output tokens across models. They should not be treated as fixed prices for every workload. The source does not provide enough detail to establish how results would change with different prompts, task mixes, or production settings. Benchmark scores and hallucination rates likewise reflect specific tests and do not settle performance or reliability for individual users.
As an affiliate, we earn on qualifying purchases.
Watch for API Access and Cost Data
The next clear developments to watch are whether Meta expands Wonder Full beyond Muse Code and whether it publishes a price or additional usage details for that model. For Wonder, customers can compare Meta’s listed token rates with their own expected input and output volumes; Artificial Analysis’s task-cost estimate offers one benchmark scenario, not a universal bill forecast.
Further independent evaluations could show whether the models’ benchmark gains carry over to other workloads and how their costs compare under consistent, practical usage. Until then, the confirmed picture is a high-scoring Full model with limited access and an API model whose measured score matches GPT-6.1 Sol at a higher benchmark-task cost.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Meta release?
Meta released Muse Wonder and Muse Wonder Full, proprietary models with a one-million-token context window. Both accept text, images, speech or audio, and video as input and return text, according to the source.
How did the models score?
Artificial Analysis scored Wonder Full at 56 and API-available Wonder at 52 on its Intelligence Index v4.3. The evaluator reported Wonder Full improved on all ten assessments compared with its index results for Muse Spark 1.3.
Can customers buy Muse Wonder Full through an API?
The source says Wonder Full is available only through Muse Code, Meta’s coding agent. It gives no per-token price for Full and does not say whether Meta plans to add API access.
What does Muse Wonder cost?
Meta’s listed API rates for Muse Wonder 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 Intelligence Index task in its test; actual costs depend on usage.
Do the benchmark results guarantee accurate answers?
No. Artificial Analysis reported hallucination rates of 23% for Wonder Full and 27% for Wonder in its evaluation. Those results are specific to the test and do not establish error rates for every task or deployment; verification remains relevant, especially for consequential uses.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
