GT Bench (Graph Theory) - Easy: leaderboard
Graph Theory Bench (Xu et al., 2026) tests whether LLMs can carry out classical graph algorithms in language on synthetic graphs given as prose, structured text, adjacency matrices or adjacency lists. The Easy subset holds connectivity and bipartiteness checks and the other tasks on sparse graphs or trees; accuracy in percent, zero-shot at temperature 0.01, higher is better.
Metric: Accuracy (%; Easy subset: connectivity and bipartiteness always, other tasks on sparse graphs or trees; zero-shot vanilla prompting at temperature 0.01, one run, over the 10,000 evaluation instances of 24 classical graph problems posed in four input representations (natural language, structured language, adjacency matrix, adjacency list); accuracy against the algorithmically verified answer). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 Mini (2025-01-31) | 98.66 | #486 |
| 2 | DeepSeek R1 | 96.32 | #467 |
| 3 | QwQ-32B | 85.62 | #909 |
| 4 | GPT-4o (2024-08-06) | 60.54 | #731 |
| 5 | Llama 3.3 70B Instruct | 58.19 | #1353 |
| 6 | Phi-4 | 53.51 | #1920 |
| 7 | GPT-4o Mini (2024-07-18) | 44.15 | #1512 |
| 8 | Llama 3.1 8B Instruct | 38.13 | #4277 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=gt-bench-graph-theory-easy · How It Works · Data refreshed daily, snapshot 2026-11-12.