GraphInstruct (Generation) - L3 Numerical Attribute Control: leaderboard
Metric: Level quality score (0-100; the paper's 0-1 score times 100) on the 150 GraphInstruct instructions of level L3, which requires target numerical attributes (density, clustering, path length); zero-shot prompting; five generations per instruction at temperature 0.7 with the InstructGraph code-style prefix; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 86.3 |
| 2 | Claude Sonnet 4 (20250514) | 84.1 |
| 3 | Qwen 3.5 122B A10B | 82.5 |
| 4 | Qwen 3.5 397B A17B | 82.4 |
| 5 | GPT-4o | 81.1 |
| 6 | Llama 3.3 70B Instruct | 80.9 |
| 7 | GPT-4.1 | 80.7 |
| 8 | Qwen 3.5 35B A3B | 80.7 |
| 9 | GPT-4o Mini | 79.3 |
| 10 | GPT-3.5 Turbo | 78.4 |
| 11 | Llama 3.1 8B Instruct | 78 |
Interactive version: theaggregate.ai/benchmark?slug=graphinstruct-generation-l3-numerical-attribute-control · How It Works · Data refreshed daily, snapshot 2026-10-07.