KG-to-Text 2026 - WebNLG17 BLEU: leaderboard
KG-to-Text study:WebNLG17 held-out generation;four prompted decoder models at temperature0,max384tokens and 3-shot cosine-retrieved demonstrations; GPT-5 Mini Minimal and open decoder thinking disabled. T5-Large and T5-Gemma2 are separately fine-tuned study checkpoints. Corpus BLEU/ChrF and mean reference metrics measure generated text; judge backbones are not scored as generators. BLEU.
Metric: BLEU. Source: arxiv.org. Saturation forecast: Around January 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 4 31B (IT) (Non-reasoning) | 53.74 | #179 (Gemma 4 31B (IT)) |
| 2 | GPT-5 Mini (Minimal) | 51.94 | #189 (GPT-5 Mini) |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kg-to-text-2026-webnlg17-bleu · How It Works · Data refreshed daily, snapshot 2026-10-24.