Padartha - Fine-Grained NER (Few-Shot): leaderboard
Padartha (2026) is named-entity recognition for classical Sanskrit on a 5,000-verse hand-corrected test set from the Mahabharata. This board scores few-shot LLMs with a two-stage pipeline on 18 fine-grained categories derived from the Nyaya-Vaisesika ontology, as micro-averaged F1 in percent; higher is better.
Metric: Micro-averaged F1 (%; entity mentions scored against 18 fine-grained ontology categories; 5,000-verse hand-corrected test set of the Mahabharata; two-stage few-shot pipeline: mention detection, then classification with 5 demonstrations retrieved by ByT5-Sanskrit embeddings). Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 61.91 | #164 |
| 2 | Gemma 4 12B (IT) | 34.28 | #693 |
| 3 | Qwen 2.5 7B Instruct | 12.77 | #2564 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=padartha-fine-grained-ner-few-shot · How It Works · Data refreshed daily, snapshot 2026-11-13.