Nejumi 4 - JBBQ (2-shot) - Accuracy: leaderboard
JBBQ, the Japanese Bias Benchmark for Question Answering, run 2-shot in Nejumi Leaderboard 4 (Weights & Biases Japan). Accuracy (%) over ambiguous and disambiguated questions; higher is better. The average absolute bias score is not published separately because it equals 1 minus the existing Nejumi 4 - ALT - Bias board.
Metric: Accuracy (%). Source: nejumi.ai. Saturation forecast: Estimated already saturated. 145 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 (Max) | 96 | #16 (Claude Opus 5) |
| 2 | GPT-5.6 Luna (Max) | 96 | #79 (GPT-5.6 Luna) |
| 3 | Claude Opus 4.7 (xHigh) | 96 | #51 (Claude Opus 4.7) |
| 4 | Claude Opus 4.8 (xHigh) | 96 | #39 (Claude Opus 4.8) |
| 5 | DeepSeek V3 (0324) | 95 | #349 |
| 6 | GPT-5.6 Sol (Max) | 95 | #22 (GPT-5.6 Sol) |
| 7 | Qwen 3 30B A3B 2507 (Thinking) | 95 | #379 |
| 8 | GLM-5.3 (Max) | 95 | #44 (GLM-5.3) |
| 9 | GPT-5.4 Mini (2026-03-17) (High) | 95 | #180 (GPT-5.4 Mini (2026-03-17)) |
| 10 | Claude Opus 4 (Non-reasoning) | 95 | #170 (Claude Opus 4) |
| 11 | O3 (2025-04-16) | 94 | #125 |
| 12 | Llama 3.1 405B Instruct FP8 | 94 | #400 |
| 13 | GLM-5.3 Flash (Max) | 94 | #77 (GLM-5.3 Flash) |
| 14 | GPT-5.6 Terra (Max) | 94 | #40 (GPT-5.6 Terra) |
| 15 | GPT-5 (2025-08-07) (High) | 94 |
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-jbbq-2-shot-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-05.