Nejumi 4 - ALT - Bias: leaderboard
Multiple-choice questions from JBBQ, the Japanese bias benchmark, covering age, disability, gender identity, physical appearance and sexual orientation. Each item gives an ambiguous or disambiguated context, a question and three options including an unknown option. The board reports one minus the average absolute bias score of the few-shot run, in percent, higher is better. Source: github.com/wandb/llm-leaderboard.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Estimated already saturated. 139 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 35B A3B | 100 | #226 |
| 2 | Claude 3.7 Sonnet | 98 | #249 |
| 3 | Qwen3.8 2.4T A95B | 97.44 | #55 |
| 4 | Qwen 3.5 397B A17B | 97.3 | #146 |
| 5 | Qwen 3.5 122B A10B | 97.14 | #173 |
| 6 | DeepSeek V3.2 | 96.57 | #205 |
| 7 | Claude Haiku 4.5 | 96.55 | #280 |
| 8 | Kimi K2.5 | 96.55 | #144 |
| 9 | DeepSeek Reasoner | 95.46 | #242 |
| 10 | Gemini 3.5 Flash | 95.24 | #60 |
| 11 | O1 (2024-12-17) | 95.24 | #149 |
| 12 | Grok 4 | 95 | #181 |
| 13 | Ling-3.0-flash-VL | 94.45 | #203 |
| 14 | Qwen 3.5 4B | 93.75 | #475 |
| 15 | GLM-4.7 Flash | 93.44 | #509 |
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-bias · How It Works · Data refreshed daily, snapshot 2026-10-20.