Open Japanese LLM v2 - JCulture-MCQ: leaderboard
From LLM-jp's Open Japanese LLM Leaderboard v2, which runs open-weight models through llm-jp-eval v2 with vLLM. JCulture-MCQ asks zero-shot four-option questions on Japanese culture that can have several correct options; metric is the set F1 between predicted and gold options. Runs evaluated before the task was added to llm-jp-eval lack it. Upstream 0-1 scores are shown as 0-100, higher is better.
Metric: Set F1 (%). Source: huggingface.co. Saturation forecast: Around 2029. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | llm-jp-4-32B-a3B-Thinking | 55.22 | #477 |
| 2 | llm-jp-4.1-33B (Thinking) | 55.15 | #256 (llm-jp-4.1-33B) |
| 3 | llm-jp-4.1-32B-a3B (Thinking) | 52.24 | #366 (llm-jp-4.1-32B-a3B) |
| 4 | GPT-OSS-120B | 49.85 | #346 |
| 5 | llm-jp-4.1-8B (Thinking) | 48.7 | #408 (llm-jp-4.1-8B) |
| 6 | llm-jp-4-8B-Thinking | 46.93 | #529 |
| 7 | GPT-OSS-20B | 34.05 | #523 |
| 8 | Qwen 3 8B (Thinking) | 21.33 | #678 (Qwen 3 8B) |
| 9 | gemma-4-26B-A4B-it (Thinking) | 19.33 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=open-japanese-llm-v2-jculture-mcq · How It Works · Data refreshed daily, snapshot 2026-11-05.