🔍 Read the full analysis: Muse Wonder Full Joins The AI Frontier; Muse Wonder Keeps Its Charge on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Artificial Analysis’ October 12 evaluation gave Meta’s Muse Wonder Full a score of 56 on its Intelligence Index, while the API-available Muse Wonder scored 52. The results show a major improvement over Muse Spark 1.3, but the source’s task-cost comparison puts Muse Wonder at $4.52 per index task, far above some similarly scored rivals; Muse Wonder Full has no public per-token price.
Artificial Analysis scored Meta’s new Muse Wonder Full at 56 on its Intelligence Index v4.3, eight points above Muse Spark 1.3, while the API model Muse Wonder scored 52. The evaluation, published October 12, puts Meta’s stronger model near leading competitors on several tests, but the source’s estimated cost per index task for Muse Wonder is $4.52—well above some rivals with similar scores.
The two models are proprietary and have a 1-million-token context window. According to the source, both accept text, images, speech or audio, and video, and return text. Their availability differs: Muse Wonder can be accessed through Meta’s API and Muse Code, while Muse Wonder Full is available only to Muse Code users. Meta’s posted 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 reported that Muse Wonder Full improved on all ten evaluations in its Intelligence Index and placed in the top three on six. Its reported results include 57% on Terminal-Bench 4.0, compared with 33% for Muse Spark 1.3, and 1,857 Elo on GDPval-AA v2.1, a test of economically useful knowledge work. The source says Claude Opus 5.5 scored 1,866 and Claude Sonnet 5.5 scored 1,838 on GDPval-AA; those results fall within the reported confidence intervals.
On the Coding Agents Index, which evaluates model-and-tool combinations, Muse Code with Wonder Full scored 68, matching Claude Code with Sonnet 5.5. Muse Code with the API model scored 63, compared with 54 for Muse Code with Spark 1.3. These are results for the tested configurations, not a direct comparison of the models running inside the same coding environment.
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 Scores, Uneven Access and Cost
The results put Meta’s new models into a more competitive position on the reported tests, particularly for coding and knowledge-work tasks. The sharp Terminal-Bench gain for Wonder Full suggests a meaningful improvement in the tested ability to complete shell-based tasks. The Coding Agents Index also indicates that users of Muse Code now have a configuration matching one of the evaluated Claude Code setups, though the index combines the model and its software harness.
Access and pricing complicate the comparison. Muse Wonder Full is the stronger model by the index score, but it is not offered through a public per-token API price in the source material. The model available through Meta’s API scores lower and, according to the source’s cost calculation, costs $4.52 per Intelligence Index task. That is a relevant consideration for organizations weighing benchmark results against expected usage costs.
The reported cost difference is not explained by output volume alone. The source says Muse Wonder generates about 54,000 output tokens per task, but consumes 7.69 million input tokens on average—more than the listed comparison models. It estimates input costs account for 76% of the $4.52 task total. Those figures describe the source’s benchmark-based calculation, not a guaranteed bill for every customer or workload.
As an affiliate, we earn on qualifying purchases.
Six Weeks After Muse Spark 1.3
The release follows Muse Spark 1.3 by about six weeks, according to the source. On the Intelligence Index, Muse Wonder Full scored 56, Muse Wonder scored 52, and Spark 1.3 scored 48. The source reports that Wonder Full improved across all ten index evaluations, while the API model also posted gains on several measures, including Terminal-Bench and AutomationBench-AA.
The source compares Muse Wonder’s score of 52 with GPT-6.1 Sol, which also scored 52, and reports task costs of $4.52 and $0.72, respectively. Gemini 4 Argon scored 53 at a reported $1.99 per task, while GPT-6 Astra scored 53 at $3.26. These cost estimates are tied to the Artificial Analysis task and methodology cited in the source; they should not be read as universal prices for every use case.
Performance is not uniform across measures. The source reports an AA-Omniscience score of 30 for Wonder Full and 25 for Wonder, compared with higher scores for several listed competitors. It also reports hallucination rates of 23% for Wonder Full and 27% for Wonder, down from 33% for Spark 1.3. The source cautions that the lower rates do not remove the need to verify answers in workflows where errors carry costs.
“The difference is the bill: one Intelligence Index task costs $4.52 on Muse Wonder and $0.72 on Sol.”
— Thorsten Meyer, ThorstenMeyerAI.com
AI model performance evaluation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Benchmarks Cannot Settle
The source does not provide a public per-token price for Muse Wonder Full, so buyers cannot use the cited figures to compare its cost per task with API-priced competitors. It also does not establish whether the task-cost estimates will match real-world spending, which can vary with prompts, workloads, and product settings. The reported scores are benchmark results and do not by themselves predict performance on every customer’s tasks.
The comparison between coding-agent scores has a further limit: the Coding Agents Index combines a model with its harness. A score for Muse Code plus Wonder Full does not show how Wonder Full would perform inside Claude Code, or how the rival models would perform in Muse Code. The source also does not give enough information here to independently assess all benchmark methods, uncertainty ranges, or the reported token-cost calculations.
AI development and benchmarking software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Pricing and Broader Availability
Meta’s next practical test is whether it expands access to Muse Wonder Full or publishes pricing that lets customers assess its cost outside Muse Code. For the API model, buyers can compare Meta’s posted token rates with their own workloads, but the source’s evaluation suggests input-token consumption may be a large part of the bill.
Further independent testing will help establish whether the benchmark gains carry over to everyday coding and knowledge-work tasks. Until more information on access, pricing, and real-world performance is available, organizations should treat the evaluation as a useful comparison point rather than a complete guide to expected results or costs.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Artificial Analysis report about Meta’s new Muse models?
It scored Muse Wonder Full at 56 and Muse Wonder at 52 on Intelligence Index v4.3. The source reports that Wonder Full improved on all ten evaluations in the index.
Can customers access Muse Wonder Full through Meta’s API?
Not according to the source. Muse Wonder Full is available only to Muse Code users, while Muse Wonder is available through Meta’s API and Muse Code.
How much does Muse Wonder cost?
Meta’s listed API price for Muse Wonder is $4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens. The source estimates one Intelligence Index task costs $4.52, but actual costs can vary by usage.
Does Muse Wonder Full have a public API price?
The source says it has no per-token price because it is available only through Muse Code. Its cost per benchmark task therefore cannot be calculated from public token pricing in the material provided.
Do the benchmark results mean the models never make mistakes?
No. The source reports hallucination rates of 23% for Wonder Full and 27% for Wonder. It says the rates improved from Spark 1.3, but they still support keeping verification steps for high-consequence work.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
