Nejumi 4 - GLP - Application Development: leaderboard
An aggregate of Nejumi 4's two application development sub-scores. Coding averages the Japanese SWE-Bench subset, JHumanEval and the MT-Bench coding questions; function calling is overall accuracy on a Japanese localisation of BFCL v3, covering single-turn, multi-turn and irrelevance detection. The board is the mean of the two in percent, higher is better. Source: github.com/wandb/llm-leaderboard.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Around June 2028. 139 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 80.38 | #12 |
| 2 | Claude Fable 5 | 70.95 | #16 |
| 3 | Claude Opus 4.8 | 70.52 | #37 |
| 4 | GPT-5.6 Sol | 70.21 | #19 |
| 5 | Claude Opus 4.7 | 69.44 | #49 |
| 6 | Claude Sonnet 4.6 | 67.95 | #90 |
| 7 | Gemma 4 31B (IT) | 67.91 | #176 |
| 8 | Qwen 3.6 Max Preview | 67.68 | #85 |
| 9 | Kimi K3 | 67.64 | #30 |
| 10 | Claude Opus 4.5 | 67.31 | #83 |
| 11 | GLM-5.2 | 66.97 | #73 |
| 12 | GPT-5.5 (2026-04-23) | 66.97 | #22 |
| 13 | GLM-5.3 Flash | 66.8 | #65 |
| 14 | Qwen3.8-Flash-Next | 66.78 | #72 |
| 15 | Gemini 3.1 Pro (Preview) | 66.76 | #58 |
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-glp-application-development · How It Works · Data refreshed daily, snapshot 2026-10-20.