How Muse Wonder Full And Muse Wonder Are Moving Into AI

🔍 Read the full analysis: How Muse Wonder Full And Muse Wonder Are Moving Into AI on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• 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 mark a notable performance step up from Muse Spark 1.3, according to an Artificial Analysis evaluation published October 12. Wonder Full scored 56 on the group’s Intelligence Index, while the API-available Wonder scored 52; the evaluation also estimated that one index task on Wonder costs $4.52, compared with $0.72 on GPT-6.1 Sol.

Both models have a one-million-token context window and accept text, images, speech or audio, and video as input, returning text. Their availability differs: Muse Wonder is offered through Meta’s API and Muse Code, the company’s coding agent, while Wonder Full is currently available only to Muse Code users. Meta’s API list price for Wonder is $4 per million input tokens and $20 per million output tokens, with cached input listed at $0.20 per million tokens.

Artificial Analysis gave Wonder Full an Intelligence Index score of 56, eight points above Muse Spark 1.3’s 48. It ranked in the top three on six of the index’s ten evaluations. On Terminal-Bench 4.0, which tests work performed in a shell, it scored 57%, up from Spark 1.3’s 33%. The evaluation also reported a 1857 Elo score on GDPval-AA, a measure of economically useful knowledge work, and a 23% hallucination rate on AA-Omniscience.

The API model scored 52 on the Intelligence Index. Artificial Analysis reported a 52% Terminal-Bench score, a 68% AutomationBench-AA score and a 27% hallucination rate for Wonder. On the same index, the evaluation placed GPT-6.1 Sol at 52 and Gemini 4 Argon at 53. The vendor’s product release and the outside evaluator’s results establish the models’ availability and measured performance; they do not by themselves show how the models perform across customers’ actual workloads.

Performance Meets API Pricing

The release gives Meta a stronger position in the market for frontier AI models and coding agents. Artificial Analysis scored Muse Code with Wonder Full at 68 in its Coding Agents Index, tying its reported score for Claude Code with Claude Sonnet 5.5. That result applies to the model-and-harness combination; it does not establish that Wonder Full would achieve the same result inside Claude Code or outperform every rival coding setup.

For buyers, the price-performance comparison is a central part of the news. Artificial Analysis estimated $4.52 per Intelligence Index task for Wonder, versus $0.72 for GPT-6.1 Sol, which received the same index score. Its analysis attributed 76% of Wonder’s task cost—$3.43—to input tokens. The evaluator said Wonder processes 7.69 million input tokens per task, compared with 6.03 million for Spark 1.3. Those are benchmark-specific estimates, not a quote for every API workload, but they suggest that token use as well as per-token rates can shape costs.

The distinction between the two releases also matters. Wonder Full is the stronger evaluated model, but users cannot buy it through a publicly listed per-token API price in the supplied material. Wonder is the model available through the API, and its score is lower. Teams weighing a switch will need to compare both capability and cost using their own prompts, task volumes and quality checks.

From Spark to Wonder

The source describes the release as arriving six weeks after Muse Spark 1.3. Artificial Analysis compared Wonder Full with 27 model configurations and reported that only Claude Opus 5.5 in its maximum configuration scored higher, at 58; Claude Sonnet 5.5 in its maximum configuration also scored 56. These are rankings within the evaluator’s stated index and configurations, rather than a universal measure of model quality.

Pricing is a sharp change from Spark 1.3’s listed rates of $1.25 per million input tokens and $4.25 per million output tokens. Wonder’s listed rates are $4 and $20, matching the figures given for Claude Opus 5.5. Artificial Analysis also reported that Wonder’s task cost was 2.8 times Spark 1.3’s in its benchmark comparison. The index task estimate reflects token consumption and evaluator methodology as well as list prices, so it should not be treated as a fixed charge for every user request.

On accuracy-related measures, the evaluator reported improvement but not market-leading results. Wonder Full’s AA-Omniscience score was 30 and Wonder’s was 25, compared with 46 for Claude Opus 5.5. Artificial Analysis reported hallucination rates of 23% for Wonder Full and 27% for Wonder, down from 33% for Spark 1.3, while Gemini 4 Argon’s rate was 15% among the comparison models scoring at least 45 on the index.

“The bill is in the input, not the output.”

— Thorsten Meyer, source article author

Costs Beyond the Benchmark

The supplied material does not include Meta’s explanation for the pricing, forecasts for API demand, or customer reports showing real-world cost and performance. Artificial Analysis’s $4.52 figure is an estimate for one Intelligence Index task; the source does not establish that all users will incur that cost for a typical request. Actual bills depend on prompts, outputs, caching and workload design.

It is also not clear whether Meta will make Muse Wonder Full available through an API, publish a per-token price for it, or expand access beyond Muse Code. The source provides no timetable for those decisions. Benchmark rankings and hallucination rates are evaluator findings, not guarantees about performance on a particular company’s tasks; users may need separate testing and verification, especially where errors carry financial or operational consequences.

Access, Pricing and Real-World Tests

The immediate next step for prospective users is to test Muse Wonder through Meta’s API or Muse Code against their own tasks and measure token use, response quality and total cost. The published list rates and Artificial Analysis estimate offer a starting point, but neither substitutes for workload-specific evaluation.

Further developments to watch include whether Meta expands Wonder Full’s availability, provides transparent pricing for it, or adjusts Wonder’s API rates. Additional independent evaluations and customer results could clarify whether the models’ benchmark gains translate into dependable performance outside the tests cited here. No such release plans or changes are confirmed in the supplied source.

Key Questions

What did Meta release?

Meta released Muse Wonder and Muse Wonder Full, proprietary models that accept text, images, audio and video as input and have a one-million-token context window. The source says both return text.

How did the models score?

Artificial Analysis scored Muse Wonder Full at 56 and Muse Wonder at 52 on its Intelligence Index v4.3. The evaluator scored Muse Spark 1.3 at 48 in the comparison.

Which model can customers access through an API?

Muse Wonder is available through Meta’s API and Muse Code. The source says Muse Wonder Full is available only to Muse Code users and gives no per-token price for it.

Why is Muse Wonder’s cost being questioned?

Artificial Analysis estimated $4.52 per index task for Wonder, compared with $0.72 for GPT-6.1 Sol, although both scored 52. The estimate reflects the benchmark and token use; it is not a universal price for every user’s workload.

What remains unknown about Muse Wonder Full?

The supplied information does not say whether Meta plans to offer Wonder Full through an API, publish a token price, or extend access beyond Muse Code. No timetable is provided.

Source: ThorstenMeyerAI.com