DeepSeek Coder V2 Lite: benchmark results
Provider: DeepSeek.
Aggregate Rating 1458 ± 44, rank #3450 of 10916 rated models, from 24 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| FormalRewardBench - Pointwise | 38.3 | Pointwise accuracy (%): the model scores the correct and the | 40 |
| LilyBench - Understanding | 56.8 | Macro average (0-1 scaled to %) over the eight Mutopia under | 33.3 |
| UniQL - Oracle | 45.44 | Execution accuracy (%) on the Oracle dialect of UniQL (1,534 | 33.3 |
| PARROT - MySQL Dialect Compatibility | 32.35 | Accuracy (%) | 28.6 |
| UniQL - Drill | 32.07 | Execution accuracy (%) on the Drill dialect of UniQL (1,534 | 25 |
| OJBench | 4.8 | Score (imported) | 20.6 |
| UniQL - Druid | 21.38 | Execution accuracy (%) on the Druid dialect of UniQL (1,534 | 16.7 |
| UniQL - MySQL | 36.57 | Execution accuracy (%) on the MySQL dialect of UniQL (1,534 | 16.7 |
| PARROT - Dialect Compatibility | 24.84 | Accuracy (%) | 14.3 |
| PARROT - Oracle Dialect Compatibility | 32.76 | Accuracy (%) | 14.3 |
| UniQL | 32.41 | Execution accuracy (%) on all 16 dialects (unweighted mean) | 8.3 |
| UniQL - All-Dialect Consistency | 4.63 | Share of the 1,534 UniQL questions (%) answered correctly (e | 8.3 |
Interactive version: theaggregate.ai/model?slug=deepseek-coder-v2-lite · How It Works · Data refreshed daily, snapshot 2026-11-08.