Muse Wonder Full Enters The AI Frontier With A Charged-Up Muse Wonder

🔍 Read the full analysis: Muse Wonder Full Enters The AI Frontier With A Charged-Up Muse Wonder on ThorstenMeyerAI.com

Before you orderOffer from Amazon

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

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that received scores of 52 and 56, respectively, in Artificial Analysis’ Intelligence Index v4.3. The stronger model is available only through Meta’s Muse Code coding agent, while Muse Wonder is available by API at prices that the evaluation estimates translate to $4.52 per benchmark task.

Artificial Analysis published its evaluation on October 12, 2026. It scored Muse Wonder Full eight points above Meta’s previous Muse Spark 1.3, which scored 48. Among 27 configurations compared in the source report, 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 guarantees of performance on every task.

Both models have a 1-million-token context window and accept text, images, speech or audio, and video as input, returning text. Meta makes Muse Wonder available through its API and inside Muse Code. Its list prices are $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. Muse Wonder Full has no stated per-token price and is accessible only to Muse Code users, leaving its cost per task unavailable for outside comparison.

The models improved on Meta’s earlier system in several reported tests. Muse Wonder Full scored 57% on Terminal-Bench 4.0, up from 33% for Muse Spark 1.3, and placed among the top three on six of the ten evaluations in the index, according to the source. Muse Wonder scored 52% on Terminal-Bench and 68% on AutomationBench-AA, up from 58% for Spark 1.3. Artificial Analysis also reported hallucination rates of 23% for Full and 27% for Wonder, compared with 33% for Spark 1.3.

Performance Comes With a Higher API Bill

The release puts Meta’s models into closer competition with leading commercial systems, but the two products pose different questions for buyers. Muse Wonder Full’s score of 56 places it near the top of the cited comparison, and its Muse Code result matches the reported score for Claude Code with Claude Sonnet 5.5. That coding-agent index evaluates the model and its software harness together; it does not establish how Wonder Full would perform inside another provider’s tool.

For API customers, the more immediate issue is cost relative to benchmark performance. Artificial Analysis estimated $4.52 per Intelligence Index task for Muse Wonder, compared with $0.72 for GPT-6.1 Sol, which also scored 52, and $1.99 for Gemini 4 Argon, which scored 53. The estimates apply to the index’s benchmark tasks, not every real-world workload, but they suggest that model scores alone may not tell buyers which service offers better value.

The source report attributes much of Wonder’s estimated task cost to its input-token use: it processed 7.69 million input tokens per task, with input accounting for $3.43, or 76%, of the reported $4.52 total. The report says Muse Wonder Full used 6.44 million input tokens per task. Actual expenses will depend on prompts, usage patterns, caching and output, so organizations would need to test their own workloads before drawing cost conclusions.

A Six-Week Step Beyond Spark

The release follows Muse Spark 1.3 by about six weeks, according to the source report. That short interval makes the reported increase in benchmark scores notable as a product update, although benchmark results alone do not establish how the underlying systems were developed or how consistently the gains will appear in practical use.

The models are not interchangeable offerings. Muse Wonder is the API product, with published token prices and a score of 52. Muse Wonder Full is the higher-scoring option but is confined to Muse Code, and Meta has not supplied a per-token price for it. The distinction matters for anyone comparing the score of the strongest model with the cost and availability of the model they can directly integrate.

Artificial Analysis also reported limits alongside the gains. Muse Wonder Full scored 30 on AA-Omniscience and Wonder scored 25, while the report listed higher scores for several competitors. The reported hallucination rates were lower than Spark 1.3’s, but remained higher than the 15% figure given for Gemini 4 Argon among models scoring at least 45 on the Intelligence Index. These measures are specific to the cited evaluations.

“The stronger model is the one you can’t buy by the token, and the model you can buy competes one tier lower.”

— Thorsten Meyer, ThorstenMeyerAI.com

Full Model Costs Remain Unpublished

Meta has not published a per-token price for Muse Wonder Full, which is available only through Muse Code. The source report therefore cannot compare its cost per task with models sold through APIs. It is also not clear from the supplied material whether Meta plans to expand Full’s availability or publish additional pricing information.

The benchmark figures come from Artificial Analysis’ evaluations; they do not settle how the models will perform across different products, prompts or production workloads. The report provides estimated costs for Intelligence Index tasks, but the supplied material does not include the full methodology or enough detail to independently calculate costs for other use cases. The extent to which the reported gains hold across customer deployments remains unestablished.

Availability and Pricing to Watch

Customers can access Muse Wonder through Meta’s API and use both models through Muse Code, with Wonder Full restricted to that coding agent. The next practical questions are whether Meta will make Full available through an API, disclose a price for it, or add further details about the models’ deployment and usage costs.

For now, organizations comparing these systems can review the published list prices for Wonder and test expected token use against their own workloads. The Artificial Analysis scores offer a standardized comparison, but they do not replace application-specific checks, particularly for tasks where an incorrect answer could have material consequences.

Key Questions

What did Meta release?

Meta released Muse Wonder and Muse Wonder Full, proprietary models with 1-million-token context windows. Both accept text, images, audio or speech, and video as input, and return text.

How did the models score?

Artificial Analysis’ Intelligence Index v4.3 gave Muse Wonder Full a score of 56 and Muse Wonder a score of 52. Its evaluation was published October 12, 2026.

How much does Muse Wonder cost?

Meta’s listed API prices 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 a $4.52 cost for one Intelligence Index task; that estimate is not a universal charge for every task.

Can customers buy Muse Wonder Full through an API?

The source report says Muse Wonder Full is available only to Muse Code users and has no stated per-token price. It does not report an API option or a public per-task cost for Full.

What remains uncertain about the release?

The supplied information does not say whether Meta will widen access to Wonder Full or publish pricing for it. Benchmark results also do not establish how either model will perform or what it will cost on an individual customer’s workload.

Source: ThorstenMeyerAI.com