SnorkelGraph: leaderboard
Snorkel AI's archived SnorkelGraph benchmark: 200 procedurally generated, expert-verified graph reasoning questions, such as minimum vertex cover, over node and edge lists given in text. Accuracy@1 against verifiable ground truth under a 16,384-token default budget; %, higher is better. Frozen roster.
Metric: Accuracy@1 (%, 200 graph reasoning questions). Source: snorkel.ai. Saturation forecast: Around February 2027. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 84.5 | #83 |
| 2 | O4 Mini | 75 | #185 |
| 3 | Grok 4 Fast (Reasoning) | 75 | #266 (Grok 4 Fast) |
| 4 | GPT-5 Mini | 72.5 | #193 |
| 5 | GPT-5 | 72 | #105 |
| 6 | O3 | 71.5 | #132 |
| 7 | O3 Mini | 71 | #282 |
| 8 | Claude Opus 4 | 64.5 | #170 |
| 9 | Grok 3 | 64 | #313 |
| 10 | GPT-4.1 | 63 | #258 |
| 11 | GPT-5 Nano | 62.5 | #439 |
| 12 | Qwen 3 235B A22B | 61.5 | #323 |
| 13 | Grok 4 | 61 | #189 |
| 14 | Gemini 2.5 Pro | 58 | #159 |
| 15 | Claude Sonnet 4 | 58 | #207 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=snorkelgraph · How It Works · Data refreshed daily, snapshot 2026-11-05.