Nejumi 4 - GLP - Coding: leaderboard
Averages three coding measures: the resolution rate on the 80-issue Japanese SWE-Bench subset; JHumanEval, a Japanese translation of HumanEval scored as the mean of its unit-test pass rate and a pylint error check; and the MT-Bench coding questions rated by an LLM judge. The board is the mean of those parts in percent where higher is better. Source: github.com/wandb/llm-leaderboard.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Around September 2027. 139 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 90.81 | #12 |
| 2 | Claude Fable 5 | 72.2 | #16 |
| 3 | Gemini 3 Flash (Preview) | 71.55 | #82 |
| 4 | Muse Spark Open | 71.21 | #54 |
| 5 | Claude Opus 4.8 | 69.12 | #37 |
| 6 | Kimi K3 | 68.47 | #30 |
| 7 | Qwen3.8-Flash-Next | 68.19 | #72 |
| 8 | Qwen3.8 2.4T A95B | 68.12 | #55 |
| 9 | Gemma 4 31B (IT) | 67.99 | #176 |
| 10 | GLM-5.3 | 67.98 | #39 |
| 11 | GPT-5.6 Sol | 67.9 | #19 |
| 12 | Hy3 | 67.9 | #93 |
| 13 | Claude Sonnet 4.6 | 67.44 | #90 |
| 14 | Gemini 3.1 Flash Lite (Preview) | 67.4 | #194 |
| 15 | GPT-5.5 (2026-04-23) | 67.34 | #22 |
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-coding · How It Works · Data refreshed daily, snapshot 2026-10-20.