GraphWalks Parents: Up to 128K
General QA · 2025-04-14
Mean node-set F1 on the GraphWalks parents task (return the parent nodes of a given node in a directed graph of hash-named nodes) aggregated over all problems with contexts up to 128K tokens, as reported by OpenAI under the label "Graphwalks parents <128k". Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.4 | 89.8 |
| GPT-5.2 | 89.0 |
| GPT-5 | 73.3 |
| o3 | 72.9 |
| GPT-4.5 | 72.6 |
| GPT-5.1 | 71.5 |
| GPT-5.4 Mini | 71.5 |
| GPT-5 Mini | 64.3 |
| GPT-4.1 Mini | 60.5 |
| o3-mini | 58.3 |
| GPT-4.1 | 58.0 |
| O4 Mini | 51.1 |
| o1 (2024-12-17) | 50.9 |
| GPT-5.4 Nano | 50.8 |
| GPT-5 Nano | 43.8 |
| GPT-4o (2024-11-20) | 35.4 |
| GPT-4o Mini | 12.6 |
| GPT-4.1 Nano | 9.4 |