GraphInstruct (Generation) - L5 Multi-Step Graph Editing: leaderboard

GraphInstruct (Generation) L5 Multi-Step Graph Editing scores zero-shot graph generation on the 50 GraphInstruct (2026) instructions whose graphs must satisfy ordered edits to a base graph. 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 50 GraphInstruct instructions of level L5, which requires ordered edits to a base graph; 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 4.684.1#179
2GPT-4o82.3#754
3Qwen 3.5 397B A17B82.1#270
4Qwen 3.5 122B A10B81.4#350
5Claude Sonnet 4 (20250514)80.6#410
6GraphInstruct DeepSeek-V3 (checkpoint unspecified)79.9
7Llama 3.3 70B Instruct79.4#1394
8Qwen 3.5 35B A3B78#494
9GPT-4.177.9#506
10GPT-3.5 Turbo75.2#3076
11GPT-4o Mini75#1680
12Llama 3.1 8B Instruct72.7#4655

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-l5-multi-step-graph-editing · How It Works · Data refreshed daily, snapshot 2026-11-22.