BLINKG - Basic - Data Reference: leaderboard
Identifying the input reference (a CSV column or an XPath expression) that each mapping reads in the basic scenario (eight atomic CSV cases over a small three-class ontology) of BLINKG (Castedo et al., Universidade de Santiago de Compostela, 2026), where an LLM fills in the knowledge-graph construction mappings for given input data and an ontology. F1 in percent against 30 gold mappings, judged by knowledge-graph experts and macro-averaged over three runs at temperature 0.3; higher is better.
Metric: F1 (%; expert-validated mappings for the source column, JSONPath or XPath each mapping reads; 30 gold task instances; eight atomic CSV cases over a three-class ontology; macro-average of three runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 89 | #262 |
| 2 | O3 | 84 | #224 |
| 3 | Llama 3.3 70B Instruct | 83 | #1353 |
| 4 | Mixtral 8x22B Instruct | 81 | #2052 |
| 5 | GPT-4o | 80 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=blinkg-basic-data-reference · How It Works · Data refreshed daily, snapshot 2026-11-12.