DNBench - Component - LLM Judge: leaderboard
DNBench tests database schema normalization with single, complex and real-world functional-dependency violations. DNB-Score combines semantic losslessness, structural agreement and logical correctness; zero-shot and few-shot configurations are separated. This series reports LLM Judge averaged over all six task/prompt configurations from 2609.11141v1, S4.T5: Component Score (0-100). Higher is better.
Metric: Component Score (0-100). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 70B | 38.3 | #580 |
| 2 | Qwen 3 30B A3B 2507 Instruct | 36.1 | #501 |
| 3 | Gemma 3 27B | 33.4 | #606 |
| 4 | Mixtral 8x7B Instruct | 22.7 | #1085 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=dnbench-component-llm-judge · How It Works · Data refreshed daily, snapshot 2026-11-07.