SchemaBridge Description - SchemaBench BLEU4 Basic: leaderboard

SchemaBridge evaluates relational-schema design, description and reconstruction. Experiment-specific cohorts retain their own populations. Graph edit distance, hallucination, structural F1, text similarity and identifier coverage are distinct measurements. Round-trip runs use a fixed GPT-4o Mini constructor with each named describer. This board reports bLEU4 (0-100). Higher is better.

Metric: BLEU4 (0-100). Source: arxiv.org. Saturation forecast: Around 2043. 5 models tracked. Updated 14 Nov 2026.

Top models

#ModelScoreOverall rank
1Llama 3.1 8B Instruct4.19#4545
2CodeLlama-13B-Instruct-hf3.79#9043
3deepseek-coder-6.7B-instruct3.02#7426
4Claude Sonnet 4.62.09#165
5GPT-5.4 Mini1.02#398

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=schemabridge-description-schemabench-bleu4-basic · How It Works · Data refreshed daily, snapshot 2026-11-16.