Nejumi 4 - ALT - Truthfulness: leaderboard
Averages two checks. JTruthfulQA is a set of Japanese questions written from scratch rather than translated, answered within fifty characters and marked right or wrong by a trained classifier; the HalluLens refusal test adds 100 questions about entities confirmed not to exist, where declining to answer is the correct behaviour. Reported in percent, higher is better. Source: github.com/wandb/llm-leaderboard.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Around January 2027. 139 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.1 | 88.5 | #140 |
| 2 | Claude Opus 4 | 88.06 | #163 |
| 3 | Claude Sonnet 4 | 86.8 | #199 |
| 4 | GPT-5.1 (2025-11-13) | 86.79 | #101 |
| 5 | Claude Opus 4.7 | 86.77 | #49 |
| 6 | Claude Fable 5 | 86.43 | #16 |
| 7 | Claude Opus 4.8 | 86 | #37 |
| 8 | Claude Opus 4.5 | 86 | #83 |
| 9 | Ornith-1.5-397B | 85.65 | |
| 10 | Qwen 3.6 Max Preview | 85.37 | #85 |
| 11 | Kimi K3 | 85.31 | #30 |
| 12 | Qwen3.8-Flash-Next | 85.31 | #72 |
| 13 | Claude Sonnet 4.5 | 85.05 | #142 |
| 14 | GLM-5.3 | 85 | #39 |
| 15 | GPT-5 | 84.9 | #98 |
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-alt-truthfulness · How It Works · Data refreshed daily, snapshot 2026-10-20.