Evaluations → Grok 4.7
Grok 4.7 commercial tested
Eighth on the extended CymraegBench leaderboard with an overall score of 82.34.
The facts
| Developer | xAI |
|---|---|
| OpenRouter ID | x-ai/grok-4.7 |
| Test pricing | $2 / $6 per million tokens (input/output) |
| Context | 500,000 tokens |
| Full run cost | $7.23 |
CymraegBench result
| Overall score | Welsh reasoning | Practical Welsh | Translation BLEU |
|---|---|---|---|
| 82.34 | 84.90 | 79.78 | 69.71 |
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.
Grok 4.7 is eighth: 0.60 points behind GPT-5.6 Sol and 0.78 points ahead of Claude Opus 4.8.
CCC Language Audit
| Subset | Score |
|---|---|
| Correction and explanation | 75.00 |
| Idiom | 93.75 |
| Terminology | 90.63 |
| Summarisation and quotation | 90.60 |
| References | 84.53 |
| Welsh prompt | 96.67 |
| English prompt | 100.00 |
| Average across the seven subsets | 90.17 |
Grok scored 78.13 for correction and 50.78 for explanation. On the references test it scored 96.88 for selecting references, 54.69 for answer content, and 100 for not inventing a source.
See the full leaderboard, CCC Language Audit, or download the summary, scores by suite, CCC dimensions and the metadata.