GraphInstruct (Generation) - L4 Domain Semantics: leaderboard

GraphInstruct (Generation) L4 Domain Semantics scores zero-shot graph generation on the 100 GraphInstruct (2026) instructions whose graphs must satisfy domain semantics (social, citation, molecular graphs). 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 100 GraphInstruct instructions of level L4, which requires domain semantics (social, citation, molecular graphs); 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: Not forecast. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.676.3#173
2GPT-4o75.1#751
3Llama 3.3 70B Instruct74.6#1385
4GPT-4o Mini74.4#1667
5GraphInstruct DeepSeek-V3 (checkpoint unspecified)74.4
6GPT-3.5 Turbo74.2#3005
7GPT-4.173.7#505
8Claude Sonnet 4 (20250514)73.7#405
9Qwen 3.5 122B A10B72.4#346
10Qwen 3.5 397B A17B72.1#264
11Qwen 3.5 35B A3B72.1#492
12Llama 3.1 8B Instruct70.3#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-l4-domain-semantics · How It Works · Data refreshed daily, snapshot 2026-11-21.