Muse Wonder commercial tested
Fourth on the extended CymraegBench leaderboard with an overall score of 86.23.
The facts
| Developer | Meta |
|---|---|
| OpenRouter ID | meta/muse-wonder |
| Test pricing | $4 / $20 per million tokens (input/output) |
| Context | 1,048,576 tokens |
| Full run cost | $17.79 |
CymraegBench result
| Overall score | Welsh reasoning | Practical Welsh | Translation BLEU |
|---|---|---|---|
| 86.23 | 86.68 | 85.78 | 71.41 |
Up to 100 cases from each of the 19 core suites were used, with reasoning_effort=low. The model completed all 2,053 cases without invalid output or unresolved infrastructure errors. The parameter seed was not sent because the model’s OpenRouter route does not support it; seed 1 was still used to select cases.
Muse Wonder is fourth: 1.18 points behind Gemini 4 Argon and 1.07 points ahead of Claude Opus 5.
CCC Language Audit
| Subset | Score |
|---|---|
| Correction and explanation | 87.66 |
| Idiom | 90.63 |
| Terminology | 100.00 |
| Summarisation and quotation | 90.60 |
| References | 90.16 |
| Welsh prompt | 65.00 |
| English prompt | 73.33 |
| Average across the seven subsets | 85.34 |
Muse Wonder scored 90.63 for correction and 62.50 for explanation. Its profile was strong on terminology and references, but weaker on following the source-selection instruction in both languages.
See the full leaderboard, CCC Language Audit, or download the summary, scores by suite, CCC dimensions and the metadata.