Nejumi 4 - JTruthfulQA: leaderboard
JTruthfulQA, a Japanese TruthfulQA-style benchmark of questions that invite false beliefs, as run in Nejumi Leaderboard 4 (Weights & Biases Japan). Free-form answers are scored by the nlp-waseda/roberta_jtruthfulqa classifier; overall score 0-100, higher is better.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Around February 2028. 145 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.1 (Thinking) | 78.01 | #146 (Claude Opus 4.1) |
| 2 | GPT-5.6 Luna (Max) | 77.15 | #79 (GPT-5.6 Luna) |
| 3 | GPT-5 (2025-08-07) (High) | 76.8 | |
| 4 | GPT-5.4 (2026-03-05) (High) | 76.8 | #63 (GPT-5.4 (2026-03-05)) |
| 5 | GPT-5.6 Sol (Max) | 76.46 | #22 (GPT-5.6 Sol) |
| 6 | GPT-5.4 (2026-03-05) (xHigh) | 76.29 | #63 (GPT-5.4 (2026-03-05)) |
| 7 | Claude Opus 4 (Non-reasoning) | 76.12 | #170 (Claude Opus 4) |
| 8 | Claude Sonnet 4 (Thinking) | 75.6 | #207 (Claude Sonnet 4) |
| 9 | DeepSeek V4.1 Flash | 75.43 | #37 |
| 10 | GPT-5.2 (2025-12-11) (xHigh) | 75.26 | |
| 11 | Gemini 2.5 Pro | 74.74 | #159 |
| 12 | GPT-5.6 Terra (Max) | 74.74 | #40 (GPT-5.6 Terra) |
| 13 | Qwen 3.6 Max (Preview) (Thinking) | 74.74 | |
| 14 | GPT-5.1 (2025-11-13) (High) | 74.57 | #107 (GPT-5.1 (2025-11-13)) |
| 15 | Grok 4.5 (High) | 74.05 | #54 (Grok 4.5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=nejumi-4-jtruthfulqa · How It Works · Data refreshed daily, snapshot 2026-11-05.