Relation Extraction on DocRED
80.2F1 ScoreDREEAM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DREEAMParadigm=Supervised (fine-tuned), Hardware=Multi-GPU†2026.04 | 80.2 | — | — | |
| ATLOPParadigm=Supervised (fine-tuned), Hardware=Multi-GPU†2026.04 | 77.8 | — | — | |
| GPT-3 few-shotParadigm=Few-shot (cloud API), Hardware=OpenAI API2026.04 | 72 | — | — | |
| SYNSYNTH (Gemma-4 Q4)Paradigm=Zero-shot (local), Hardware=1× RTX 30902026.04 | 70.2 | — | — | |
| SMADE-IEBackbone=GPT-3.5-Turbo-0125, Prompting Strategy=multi-agent debate2026.06 | 30.78 | 26.99 | — | |
| AEiOBackbone=GPT-3.5-Turbo-0125, Prompting Strategy=monolithic prompting2026.06 | 30.32 | 24.84 | — | |
| GPT-3 zero-shotParadigm=Zero-shot (cloud API), Hardware=OpenAI API2026.04 | 30 | — | — | |
| ChatGPT zero-shotParadigm=Zero-shot (cloud API), Hardware=OpenAI API2026.04 | 25 | — | — | |
| One-StepBackbone=GPT-3.5-Turbo-0125, Prompting Strategy=each-type prompting2026.06 | 22.15 | 18.41 | — | |
| G&OBackbone=GPT-3.5-Turbo-0125, Prompting Strategy=each-type prompting2026.06 | 19.69 | 16.71 | — | |
| CROSSAGENTIEBackbone=GPT-3.5-Turbo-0125, Prompting Strategy=multi-agent debate2026.06 | 13.72 | 11.49 | — | |
| GLiNER-RelexEntities=predicted2026.05 | — | 31.3 | 25.6 | |
| GLiNER2Entities=predicted2026.05 | — | 11.7 | 17.8 | |
| GLiRELEntities=gold2026.05 | — | 2.4 | 8.1 | |
| GPT-5-miniEntities=prompted2026.05 | — | 18.6 | 22.1 |