GABench - Graph Retrieval - Edge Level (Numerical-Attribute Graphs): leaderboard
Metric: Task success rate (%; share of tasks whose final answer a DeepSeek-V4-Flash judge marks consistent with the verified ground truth; edge-level graph retrieval (node attributes, neighbourhoods, edges, subgraphs and graph statistics) with its 10 retrieval tools on numerical-attribute graphs (DGraph-Fin, Pokec, NBA, Credit and Bail, 200 generated tasks per dataset and category); LLM agent in the OpenClaw harness, one run). Source: arxiv.org. Saturation forecast: Around October 2027. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 Mini | 70.59 |
| 2 | DeepSeek V4 Pro | 59.66 |
| 3 | Qwen 3 235B A22B | 6.66 |
| 4 | GLM-5-Turbo | 5 |
| 5 | GLM-4.5 Flash | 4.73 |
| 6 | Gemini 3 Flash | 0.8 |
Interactive version: theaggregate.ai/benchmark?slug=gabench-graph-retrieval-edge-level-numerical-attribute-graphs · How It Works · Data refreshed daily, snapshot 2026-09-29.