GraphInstruct (Generation) - L3 Numerical Attribute Control: leaderboard

GraphInstruct (Generation) L3 Numerical Attribute Control scores zero-shot graph generation on the 150 GraphInstruct (2026) instructions whose graphs must satisfy target numerical attributes (density, clustering, path length). The level quality score, 0 to 100 and higher is better, is computed by code from validity, distribution match, reference similarity and constraint satisfaction.

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 February 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.686.3#173
2Claude Sonnet 4 (20250514)84.1#405
3GraphInstruct DeepSeek-V3 (checkpoint unspecified)83.7
4Qwen 3.5 122B A10B82.5#346
5Qwen 3.5 397B A17B82.4#264
6GPT-4o81.1#751
7Llama 3.3 70B Instruct80.9#1385
8GPT-4.180.7#505
9Qwen 3.5 35B A3B80.7#492
10GPT-4o Mini79.3#1667
11GPT-3.5 Turbo78.4#3005
12Llama 3.1 8B Instruct78#4600

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=graphinstruct-generation-l3-numerical-attribute-control · How It Works · Data refreshed daily, snapshot 2026-11-21.