Muse Wonder Full Arrives On AI’s Frontier As Muse Wonder Charges Too

🔍 Read the full analysis: Muse Wonder Full Arrives On AI’s Frontier As Muse Wonder Charges Too on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Meta has released Muse Wonder and Muse Wonder Full, two proprietary AI models that received strong scores in an evaluation published by Artificial Analysis on October 12, 2026. Wonder Full scored 56 on the Intelligence Index, while Wonder, the version available through Meta’s API, scored 52; the evaluation’s estimated cost per task for Wonder was $4.52, higher than several similarly scoring competitors.

Artificial Analysis said Muse Wonder Full improved on all 10 evaluations in its Intelligence Index v4.3 compared with Meta’s Muse Spark 1.3. The report places Wonder Full behind Claude Opus 5.5, which scored 58, and level with Claude Sonnet 5.5 at 56 among the configurations it compared. These are third-party benchmark results, not a guarantee of performance in every use.

The models differ in access. Muse Wonder is available through Meta’s API and in Muse Code, the company’s coding agent. Its listed prices are $4 per million input tokens and $20 per million output tokens, with cached input priced at $0.20 per million tokens. Muse Wonder Full is available only to Muse Code users and has no per-token price listed, leaving its cost per task unclear to outside users.

The evaluation reports notable gains on particular tests: Wonder Full scored 57% on Terminal-Bench 4.0, up from 33% for Muse Spark 1.3, and ranked third on several other evaluations. The source also reports a 23% hallucination rate for Wonder Full, compared with 33% for Spark 1.3. Those figures reflect the stated benchmark methods and should not be read as universal rates across real-world use.

Strong Scores Meet High API Costs

The release matters because it puts Meta’s models into the same benchmark conversation as leading commercial systems while raising a practical question for developers: how much performance costs to use. Artificial Analysis estimates one Intelligence Index task with Muse Wonder at $4.52. In its comparison, GPT-6.1 Sol scored the same 52 for $0.72, while Gemini 4 Argon scored 53 for $1.99. The reported figures make Wonder substantially more expensive on that test, although task costs depend on the evaluation’s workload and token use.

The distinction between the two Muse products also affects buyers’ choices. The higher-scoring Wonder Full is restricted to Muse Code, while the model developers can call through Meta’s API is the lower-scoring Wonder. This limits direct price and access comparisons for Wonder Full and means an API customer cannot assume the top benchmark result is available as a separately priced model.

Artificial Analysis also reports a Coding Agents Index score of 68 for Muse Code plus Wonder Full, tying its score for Claude Code plus Claude Sonnet 5.5. That index measures the model together with its software harness. It indicates a competitive packaged coding tool in this test, but does not establish how Wonder Full would perform inside another coding agent or on every programming task.

Two Models, Different Access

The source dates the release to about six weeks after Muse Spark 1.3. Both new models are described as proprietary, with a one-million-token context window. They accept text, images, speech or audio, and video as input, and return text. Meta’s stated product access separates the higher-scoring Wonder Full from the API model used for the report’s cost comparison.

Artificial Analysis’s Intelligence Index combines results from multiple evaluations. Its reported results include Wonder Full’s 1,857 Elo on GDPval-AA, which the source places close to Claude Opus 5.5 at 1,866 and Claude Sonnet 5.5 at 1,838. On AA-Omniscience, however, the report gives Wonder Full a score of 30 and Wonder 25, below several listed competitors. The pattern is one of strong but uneven results across tasks, rather than a single score establishing broad superiority.

The source also highlights token use as a contributor to the estimated API bill. It says Wonder used 7.69 million input tokens for an Intelligence Index task, compared with 6.03 million for Spark 1.3, and generated about 54,000 output tokens. The supplied material attributes $3.43, or 76% of Wonder’s $4.52 task cost, to input. These are evaluation-specific measurements, not fixed charges for every API request.

Costs and Real-World Results

The source does not provide a per-task price for Muse Wonder Full, which is limited to Muse Code. It is also unclear whether Meta will offer that model through its API or publish enough pricing and usage information for customers to compare its costs directly with Wonder and competing services.

The reported $4.52 cost for Muse Wonder is tied to the Intelligence Index workload and its measured token consumption. It does not establish what a typical customer will pay for a specific application. The supplied material also does not include a detailed account of the evaluation’s uncertainty ranges for every result, or independent testing of the models in production settings.

Benchmark scores and reported hallucination rates cannot settle how reliably the models will perform across different users, prompts, and tasks. Readers should treat the figures as results under the stated evaluation, not assurances that a model will achieve the same accuracy or cost in a particular workflow.

Pricing and Access Will Matter

The next practical test will be how developers and businesses respond to Muse Wonder’s API pricing and its performance on their own workloads. Customers can compare actual token consumption, output quality, and verification needs against other available models rather than relying only on a benchmark task estimate.

Further details from Meta on Wonder Full’s availability, pricing, and intended access could clarify whether the stronger model will remain confined to Muse Code. Additional independent evaluations may also show whether the reported gains hold across a wider range of tasks. The supplied source does not state when Meta might make either change or publish further results.

Key Questions

What are Muse Wonder and Muse Wonder Full?

They are two proprietary Meta models. The source says both accept text, image, audio or speech, and video inputs, and return text. Wonder is available through Meta’s API and Muse Code; Wonder Full is available only through Muse Code.

How did the models score in the reported evaluation?

Artificial Analysis gave Muse Wonder Full a score of 56 and Muse Wonder a score of 52 on its Intelligence Index v4.3. These are benchmark results and do not establish performance on every task.

How much does Muse Wonder cost?

The listed API price is $4 per million input tokens and $20 per million output tokens, with cached input at $0.20 per million. Artificial Analysis estimated $4.52 for a task in its Intelligence Index, but that is specific to the evaluation’s measured token use.

Is Muse Wonder Full available through the API?

Not according to the supplied source. It says Wonder Full is available only to Muse Code users and has no listed per-token price, so its per-task cost cannot be compared from the reported pricing information.

Does the evaluation show that Muse Wonder is better than competing models?

No single score establishes that. Wonder Full matched some leading systems on reported measures and ranked highly on several tests, while the results also show weaker relative performance on other measures. Benchmark rankings depend on the task and evaluation method, and real-world performance may differ.

Source: ThorstenMeyerAI.com