Doc2DB-Bench - Entity Tables: leaderboard
Cell-level F1 of the entity tables a model extracts from long documents in Doc2DB-Bench (2026). Predicted and gold rows are aligned by globally sorting candidate pairs by similarity and greedily assigning one-to-one matches. Percent; higher is better.
Metric: Entity-level cell F1 (%; entity tables extracted from the document; 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 February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 84.93 | #102 |
| 2 | GPT-5.4 | 80.09 | #123 |
| 3 | Gemini 2.5 Pro | 78.13 | #266 |
| 4 | Qwen 3 Max | 75.77 | #403 |
| 5 | Gemini 2.5 Flash | 74.21 | #453 |
| 6 | GPT-4o | 70.3 | #741 |
| 7 | DeepSeek V4 Flash | 67.3 | #258 |
| 8 | Qwen 2.5 72B Instruct | 45.03 | #1092 |
| 9 | Qwen 2.5 14B Instruct | 44.07 | #1732 |
| 10 | Llama 3.1 70B Instruct | 17.25 | #1546 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=doc2db-bench-entity-tables · How It Works · Data refreshed daily, snapshot 2026-11-13.