SnorkelSequences: leaderboard
Snorkel AI's archived SnorkelSequences benchmark: 250 procedurally generated compositional questions that apply operators, conditions and transforms to numeric sequences, with programmatically verifiable answers and no LLM judge. Accuracy@1; %, higher is better. Frozen roster.
Metric: Accuracy@1 (%, 250 compositional sequence questions). Source: snorkel.ai. Saturation forecast: Around February 2027. 32 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 77.6 | #105 |
| 2 | GPT-5 Mini | 77.6 | #193 |
| 3 | GPT-5 Nano | 72 | #439 |
| 4 | GPT-5.4 | 71.6 | #83 |
| 5 | O3 Mini | 71.2 | #282 |
| 6 | Gemini 2.5 Flash | 70.8 | #255 |
| 7 | Claude Sonnet 4 | 70.4 | #207 |
| 8 | Grok 4 Fast (Reasoning) | 70.2 | #266 (Grok 4 Fast) |
| 9 | O4 Mini | 68.8 | #185 |
| 10 | Gemini 2.5 Pro | 66 | #159 |
| 11 | Claude Opus 4 | 65.6 | #170 |
| 12 | O3 | 65.2 | #132 |
| 13 | Grok 4 | 63.2 | #189 |
| 14 | Llama 4 Maverick | 62 | #486 |
| 15 | Nova Premier | 51.8 | #447 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=snorkelsequences · How It Works · Data refreshed daily, snapshot 2026-11-05.