Doc2DB-Bench - Transitive Inference: leaderboard

Metric: Cell-level F1 (%; relation-table cells annotated with the Transitive Inference capability of the inter-table taxonomy (Section 5.3); 203 synthesized long-document instances over 42 database schemas from BIRD and Spider in seven domain groups; identical prompts, greedy decoding at temperature 0; cells aligned by global maximum-weight tuple matching, a cell matching on exact numeric equality or at least 90 percent string similarity). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.

Top models

#ModelScore
1GPT-5.478.47
2Gemini 2.5 Pro70.82
3GPT-4o65.51
4Claude Opus 4.659.48
5Llama 3.1 70B Instruct57.83
6DeepSeek V4 Flash57.35
7Qwen 3 Max53.78
8Qwen 2.5 14B Instruct40.8

Interactive version: theaggregate.ai/benchmark?slug=doc2db-bench-transitive-inference · How It Works · Data refreshed daily, snapshot 2026-09-29.