Compare / DeepSeek V4 Pro 0423 vs Ministral 3 14B 2512
DeepSeek V4 Pro 0423 vs Ministral 3 14B 2512
14 results from 3 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | DeepSeek V4 Pro 0423 | Ministral 3 14B 2512 |
|---|---|---|---|
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 68.8 | 14.4 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 36.0 | 6.00 |
| Artificial Analysisreported | GPQA Diamond% of questions | 92.8% | 57.2% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 41.0% | 4.6% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 80.3% | 26.3% |
| Artificial Analysisreported | SciCode% of problems | 51.0% | 23.8% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 78.7% | 9.7% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 14.1% | 0.0% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 39.6% | 6.6% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 34.0% | 24.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.36 | 0.32 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 68.4 | 31.4 |
| Vectara Hallucination Leaderboardreported | Answer rate% of documents | 97.2% | 99.6% |
| Vectara Hallucination Leaderboardreported | Hallucination rate% of summaries | 8.6% | 19.4% |
Bold green marks the better value on that metric. Where a source reports ranges that overlap, the difference may not be meaningful; see the benchmark page for ranges.