Nejumi 4 - GLP - Abstract Reasoning: leaderboard
Grid puzzles from ARC-AGI-1 and ARC-AGI-2: the model sees a few demonstration input and output grids, infers the transformation and writes the output grid for a new input. Nejumi uses 50 tasks from each public evaluation set, with Japanese system prompts and the original numeric grids, and averages the two accuracies in percent where higher is better. Source: github.com/wandb/llm-leaderboard.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Around January 2027. 139 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Terra | 98 | #41 |
| 2 | GPT-5.6 Sol | 97 | #19 |
| 3 | Claude Opus 5 | 96 | #12 |
| 4 | Claude Fable 5 | 93 | #16 |
| 5 | GPT-5.5 (2026-04-23) | 93 | #22 |
| 6 | GPT-5.4 (2026-03-05) | 91 | #59 |
| 7 | Qwen3.8 2.4T A95B | 90 | #55 |
| 8 | Claude Opus 4.8 | 87 | #37 |
| 9 | Claude Opus 4.6 | 85 | #63 |
| 10 | Gemini 3.6 Flash | 84 | #51 |
| 11 | Gemini 3.1 Pro (Preview) | 83 | #58 |
| 12 | Claude Opus 4.7 | 83 | #49 |
| 13 | Hy4 preview | 83 | #38 |
| 14 | DeepSeek V4.1 Flash | 81 | #36 |
| 15 | Grok 4.20 0309 (Reasoning) | 81 | #112 (Grok 4.20 0309) |
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-abstract-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.