SchemaBridge Direct Gold Text - RSchema Hallucination: 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 four-dimensional hallucination ratio (0-1). Lower is better.

Metric: Four-dimensional hallucination ratio (0-1). Source: arxiv.org. Saturation forecast: Around 2032. 8 models tracked. Updated 14 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1deepseek-coder-6.7b-instructSchemaBridge run: Few-shot0.582
2CodeLlama-13b-Instruct-hfSchemaBridge run: Few-shot0.589
3deepseek-coder-6.7B-instruct0.595#7426
4GPT-5.4 Mini0.605#398
5Claude Sonnet 4.60.635#165
6Meta-Llama-3.1-8B-InstructSchemaBridge run: Few-shot0.638
7CodeLlama-13B-Instruct-hf0.669#9043
8Llama 3.1 8B Instruct0.696#4545

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

Interactive version: theaggregate.ai/benchmark?slug=schemabridge-direct-gold-text-rschema-hallucination · How It Works · Data refreshed daily, snapshot 2026-11-16.