Mistral Large 4 commercial tested
Ranked 23rd on the extended CymraegBench leaderboard with an overall score of 59.93.
The facts
| Developer | Mistral AI |
|---|---|
| OpenRouter ID | mistralai/mistral-large-4-0 |
| Test pricing | $0.68 / $2.09 per million tokens (input/output) |
| Context | 1,048,576 tokens |
| Full run cost | $0.23 |
CymraegBench result
| Overall score | Welsh reasoning | Practical Welsh | Translation BLEU |
|---|---|---|---|
| 59.93 | 66.08 | 53.78 | 59.28 |
Up to 100 cases from each of the 19 core suites were used. On this route, OpenRouter offers only reasoning_effort=high or none ; the setting none was used as the lowest available option. The model completed all 2,053 cases; three transient API errors were successfully retried.
Mistral Large 4 ranks 23rd: 1.70 points behind DeepSeek V4-Flash-0731 and 1.80 points ahead of Grok 4.20.
CCC Language Audit
| Subset | Score |
|---|---|
| Correction and explanation | 59.30 |
| Idiom | 84.38 |
| Terminology | 53.13 |
| Summarisation and quotation | 96.07 |
| References | 73.75 |
| Welsh prompt | 93.33 |
| English prompt | 95.83 |
| Average across the seven subsets | 79.40 |
Mistral Large 4 scored 71.88 for correction and 42.97 for explanation. Summarisation and instruction following were its main strengths in the audit; terminology was the weakest test.
See the full leaderboard, CCC Language Audit, or download the summary, scores by suite, CCC dimensions and the metadata.