Nejumi 4 - GLP - Mathematical Reasoning: leaderboard
Averages the MAWPS and MGSM math word problem sets from the jaster harness, each taken as the mean of its 0-shot and 2-shot runs and graded by exact match on the final answer, with the MT-Bench math questions rated by an LLM judge. Reported on a 0 to 100 scale where 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 | Claude Opus 5 | 98.17 | #12 |
| 2 | Claude Fable 5 | 98 | #16 |
| 3 | Claude Opus 4.8 | 97.5 | #37 |
| 4 | Claude Opus 4.7 | 97.33 | #49 |
| 5 | Gemini 3.6 Flash | 97.33 | #51 |
| 6 | Gemini 3 Flash (Preview) | 97.17 | #82 |
| 7 | Gemini 3.5 Flash | 97.17 | #60 |
| 8 | GPT-5.1 (2025-11-13) | 97.17 | #101 |
| 9 | Gemini 3 Pro (Preview) | 97 | #66 |
| 10 | Gemini 3.1 Pro (Preview) | 96.83 | #58 |
| 11 | DeepSeek V4.1 Flash | 96.83 | #36 |
| 12 | GPT-5.4 (2026-03-05) | 96.83 | #59 |
| 13 | Hy4 preview | 96.83 | #38 |
| 14 | GPT-5.2 | 96.67 | #110 |
| 15 | Qwen3.8-Flash-Next | 96.67 | #72 |
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-mathematical-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.