GraphInstruct (Generation) - L0 Format Generation: leaderboard

GraphInstruct (Generation) L0 Format Generation scores zero-shot graph generation on the 100 GraphInstruct (2026) instructions whose graphs must satisfy format generation (node and edge counts, syntax). 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 L0, which requires format generation (node and edge counts, syntax); 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: Estimated already saturated. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.695.8#179
2Qwen 3.5 397B A17B95.2#270
3Qwen 3.5 122B A10B94.7#350
4GraphInstruct DeepSeek-V3 (checkpoint unspecified)94.5
5GPT-4o94.2#754
6Claude Sonnet 4 (20250514)93.9#410
7Qwen 3.5 35B A3B92.4#494
8GPT-4.192.3#506
9Llama 3.3 70B Instruct90.5#1394
10Llama 3.1 8B Instruct89.6#4655
11GPT-4o Mini85.2#1680
12GPT-3.5 Turbo85#3076

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=graphinstruct-generation-l0-format-generation · How It Works · Data refreshed daily, snapshot 2026-11-22.