llm-jp-3.1-1.8B-instruct4: benchmark results
National Institute of Informatics' llm-jp-3.1-1.8b-instruct4, an instruction-tuned 1.8B model in the LLM-jp-3.1 series, adding mid-training (instruction pre-training) on top of the original LLM-jp-3 recipe. Provider: National Institute of Informatics. Released 2025-05-27. Access: Open.
Aggregate Rating 1387 ± 27, rank #1249 of 1635 rated models, from 85 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Japanese LLM - FA | 53.84 | Score (%) | 99.8 |
| Open Japanese LLM - Janli Exact Match | 100 | Score (%) | 99.8 |
| Open Japanese LLM - Wikicorpus J TO E Bert Score EN F1 | 92.15 | Score (%) | 99.6 |
| Open Japanese LLM - Jamp Exact Match | 93.68 | Score (%) | 99.5 |
| Open Japanese LLM - Jsick Exact Match | 90.05 | Score (%) | 99.5 |
| Open Japanese LLM - Kuci Exact Match | 88.85 | Score (%) | 99.5 |
| Open Japanese LLM - NLI | 92.44 | Score (%) | 99.5 |
| Open Japanese LLM - Wiki Dependency SET F1 | 85.65 | Score (%) | 99.5 |
| Open Japanese LLM - Wiki NER SET F1 | 20.35 | Score (%) | 99.5 |
| Open Japanese LLM - Wiki PAS SET F1 | 43.76 | Score (%) | 99.5 |
| Open Japanese LLM - Wikicorpus E TO J Bleu JA | 20.47 | Score (%) | 99.5 |
| Open Japanese LLM - Wikicorpus J TO E Bleu EN | 14.65 | Score (%) | 99.5 |
Interactive version: theaggregate.ai/model?slug=llm-jp-3-1-1-8b-instruct4 · How It Works · Data refreshed daily, snapshot 2026-10-15.